6.

1 Suprasegmental Phonology 187

\Vjthin the suprasegmental unit (here, the syllable). A segmental analysis, on
the other hand, would attempt to assign an underlying [+nasal] feature
spec::ification to one segment within each suprasegmental unit and then
provide a rule by which neighboring segments assimilate to that feature
specification. In the underlying form /ba/, nasality is assigned to underlying
vowels. A rule is therefore required to nasalize oral consonants in the context
of a following nasalized vowel, as seen below:
C-+ [+nasal]/- [+nasal]
v
An equally plausible segmental analysis would recognize the underlying

SUPRftSEGN\ENTftL form fma/, where nasality is assigned to the consonant. In this case a rule is
needed to nasalize an oral vowel following a nasal consonant, as seen below:

PHONOLOGY V-+ [+nasal] I [+nasal]_

6.0 The Study of Suprasegmentals
Much of the current research in phonology has focused on units
larger than the segment. Stress, tone, and duration (vowel and consonant
length) are often claimed to be properties of suprasegmental units such as
the syllable or word, while vowel harmony and nasalization are also some-
times i?cluded under this heading (Firth, 1948; Robins, 1957b). That is,
prosodic features such as those just mentioned are best seen as extending
over units which can encompass more than one segment. For example, many
languages require that all segments within a syllable agree in nasality. Thus,
a CV sequence consisting of a voiced labial stop and a low back vowel can
be realized phonetically as [ba] or [rna], but not as *[bii] or *[rna]. As seen
below,
SUPRASEGMENTAL ANALYSIS SEGMENTAL ANALYSIS
[ba] /baf /ba/
[rna] /ba/N /bii/ or /rna/
in a suprasegmental or prosodic analysis a nasal exponent can be factored
out which, by a "mapping rule" (Leben, 1973a,b), is assigned to each segment
186
c
While both of these rules are "natural" in the sense discussed in Chapter 5,
it is also possible to analyze nasalization as a suprasegmental property, as
discussed in 6.3.2.
The issue of whether certain phonological phenomena should be analyzed
segmentally or suprasegmentally (that is, prosodically, in the British terminol-
ogy) has been of concern to phonologists. In addition, many of the central
issues in phonological theory have been argued on the basis of suprasegmental
phenomena-in particular, stress, but also tone, duration, vowel harmony,
and nasalization. It is thus appropriate that the last chapter of this book
address itself to questions of suprasegmentality.
6.1 Suprasegmental Units
In the preceding section it was seen that the same phonological data
might be analyzed segmentally or suprasegmentally, depending on one's
particular theory of phonology. We find not only this potential disagreement
among phonologists, but also a second disagreement about which supra-
segmental units are required in phonology. The term "suprasegmental" is
used to refer to both phonological and grammatical units larger than the
segment. In both categories there is disagreement.
6.1.1 Phonological Units
Phonological suprasegmentals are those which are defined in terms
of the sound segments of which they are comprised. While the boundaries of
 188 Suprasegmental Phonology 6.1
these units are sometimes affected by grammatical considerations,
logical units do not in themselves have a grammatical basis or function.
6.1.1.1 The Syllable By far the most widely discussed pn;Dncnoll~C!:Il
suprasegmental is the syllable. While the study of the syllable has a
uninterrupted history (see Allen, 1973 and Pulgram, 1970 for rei,erence~\;;;
there are typically three questions which arise in this context: (1) how
one define the syllable? (2) how does one determine syllable uv ...... ,.a.lllt'<>'l
and (3) is the syllable a necessary concept?
One can readily divide the Shona word murume 'man' into the
syllables mu, ru, and me. Since this word has a evevev structure
division into three phonological parts creates three sequences of CV'
optimal syllable structure. As stated by Malmberg (1963:129), "A v,',,n..,, .
consisting of a consonant plus a vowel represents the most primitive, . principle of maximal open syllabicity, (2) a principle of minimal coda and
without doubt historically the oldest, of all syllable types, the only one which
is general in all languages." Whenever languages have syllable types other
than ev, complications arise in the exact determination of syllable bound
aries. For instance, should a word with the structure eveev be syllabified
as eV$eev or as eVe$eV?
6.1.1.1.1 Defining the Syllable Before anything can be determined
about syllable division, it is necessary to establish some idea of what is meant
by the syllable. In particular, is the syllable a phonological unit, a phonetic
unit, both, or neither? Most phonologists, to the extent that they have
accepted it, attempt to deal with the syllable as a phonological unit. As such
words and larger utterances can be syllabified on the basis of the phonotacti~
(or sequential) constraints of a given language, subject to certain universaf
tendencies.
The syllable consists of three phonetic parts: (1) the onset, (2) the peak
or nucleus, and (3) the coda. In a syllable such as man, fmf is the onset, /ref is
the peak, and /n/ is the coda. For phonological purposes, however, only a
sing!~ ~ivision is relevant, namely between (1) the onset and (2) the core,
cons1stmg of the phonetic peak and coda combined. This analysis of the
syllable (see Pike and Pike, 1947), as represented below,
syllable
onset~ore
~coda
peak
divides a eve syllable into e-Ve rather than ev-e or e-V-e. In so doing,
we are able to capture the important distinction between open and closed
syllables, as well as between heavy and light syllables (which will be discussed
in 6.2.1.2.2). An open syllable ends in a vowel, while a closed syllable is
"checked" or "arrested" by a consonant. A ev syllable thus has a core
6.1 Suprasegmental Phonology 189
with a zero coda, while a eve syllable has a core with a V peak and a e coda.
The initial consonant onset is irrelevant in determining the phonological
properties of a syllable.
The basic assumption in phonological approaches to the syllable is that
there is an intimate relationship between word structure and syllable structure.
Thus, ideally, the same sequential constraints which operate at the beginning
of a word should be operative at the beginning of a syllable, even if this
syllable is word-internal. Similarly, the same sequential constraints which
operate at the end of a word should be operative at the end of a syllable.
Attempts to provide universal principles for determining syllable structure
are represented by Pulgram (1970) and Hooper (1972). Recognizing a parallel
between word structure and syllable structure, Pulgram proposes (1) a
maximal onset, and (3) a principle of the irregular coda.
By the first principle, a syllable boundary is inserted after every vowel (or
diphthong) of a word. 1 Thus, words such as rooster and master are syllabified
as roo$ster /ru$st<Jr/ and ma$ster /mre$st<Jr/, so as to make the first syllable
open. A problem arises in the form ma$ster, however, since the principle of
maximal open syllabicity creates a sequence which violates a sequential
constraint in English by which the lax vowels /I, e, u, o, ref are disallowed in
word-final position. Since ma$ster contains the vowel tre/, which does not
occur word-finally, it must be resyllabified by the second principle to yield
mas$ter. As stated by Pulgram (1970), "If a syllable cannot be kept open
because its vowel does not occur in word-final position, then as many
consonants as necessary-but no more-to provide the syllable with a
permissible coda, thereby removing the vowel from the syllable-final position,
must be detached from the onset of the next syllable and transferred to the
preceding syllable" (p. 48).
A similarly motivated readjustment must occur in a second set of cir-
cumstances: " If the syllable cannot be kept open because the consonant or
consonants that would form the onset of the next syllable do not occur in
word-initial position, then as. many consonants as necessary-but no more--
to reduce the onset to a permissible word-initial shape must be detached
from it and transferred to the preceding syllable as coda, thus closing the
syllable" (p. 50). Thus, while employ would be syllabified e$mploy by the
principle of maximal open syllabicity, this would create a syllable-initial
mpl sequence, which cannot occur word-initially. Thus, the m must be sent
back to the first syllable to yield em$ploy, where each syllable now meets the
word-structure constraints of English.
Pulgram's final principle is stated as follows: "If the necessary transfer
1
Pulgram actuaily refers to a concept of the "section," which is the domain of syilabifica-
tion.
 190 Suprasegmental Phonology 6.1
from syllable-initial to syllable-final position leads to an inadmissible
final group of consonants, then the burden of irregularity must be
the coda rather than the following onset" (p. 51). The example which
gives is Spanish transcribir, which, according to the first principle,
be syllabified tra$nscribir. Since nscr (where c = fk/) cannot occur
initially in Spanish, the n must be transferred to the first syllable to
tran$scribir. However, scr still is not an acceptable word-initial
and so the s must also be transferred to the first syllable, yielding
The result, however, is that the first syllable now has a final sequence
which cannot occur word-finally in Spanish. The syllabification trunx.,"'"'"'-
is preferable to tran$scribir, since the coda is more capable than the ~ ..U..,ur
allowing violations of word-structure constraints.
This last principle of Pulgram's correlates with the observation that
languages allow longer consonant sequences at the ends of syllables than at
the beginning of syllables. 2 Thus Berber allows CC$ sequences but not $CC
sequences (Jilali Saib, personal communication). However, the fact that ns
can end a syllable but cannot end a word in Spanish illustrates the possibility
t~at syllable-structure and word-structure constraints may occasionally
d1ffer. Vennemann (l972a:l3) claims that in some dialects of German radle
(from radele) '(I) go by bicycle' is syllabified ra$dle (pronounced [ra:dl~]),
despite the fact that German does not allow word-initial dl sequences.
Other dialects syllabify this word as rad$le, in which case it must be pro-;
nounced with syllable-final devoicing, that is, as [ ra: tl~].
While such phonotactic approaches to the syllable define syllable breaks
in terms of phonological constraints, less is said about how a word would
be carefully divided into parts if spoken slowly. A word such as bedroom
must be syllabified bed$room, because jej is not a permissible word-final
vowel in English. However, as pointed out by Ferguson (1962:373), many
speakers of English differentiate between bedroom 'the room in which one
sleeps' and bedroom 'space for a bed.' While the second is uniformly pro-
nounced [bed$ruwm], with correct syllabification following Pulgram's
principles, the first is often pronounced [be$druwm], and even [be$jruwm].
In these last two transcriptions, the syllable boundary represents the point at
which a pause could conceivably be taken. What we observe is that the d of
bed might be assigned to the following syllable, and that the syllable-initial
dr sequence might even undergo affrication to [Jr], exactly as observed in
word-initial dr sequences (as in [dres] or [jres] 'dress').
Pulgram would call be$droom a "nexus," while we might propose recog-
nizing the difference between these two forms as one between different internal
boundaries, that is, bed+room vs. bed#room (see 6.1.2.2). If we attempted to
define the syllable breaks according to whether or not dr undergoes syllable-
2
On the other hand, certain languages, such as Ewe, have CLV (consonant-liquid-vowel)
but not *CVL.
6.1 Suprasegmental Phonology 191
affrication, we would say that these two words syllabify differently.
Since a word such as excedrin with no boundary syllabifies as [ ek$se$drm]
or [sk$se$jnn], it would appear to be the general case that VdrV syllabifies
:as V$drV-unless blocked by a strong boundary(#). However, the criterion
of affrication is in direct conflict with phonological syllabification, according
to which lax vowels such as fej cannot end syllables.
An alternative treatment of the syllabification of VCV sequences, where
the first vowel is lax, is to assign the intervocalic consonant simultaneously to
both syllables. Thus, the words bacon and beckon would be syllabified as
$
[be$k~n] and [bebn]. In beckon, the syllable boundary comes within the
jk/, which is sometimes claimed to be long or geminate. In this analysis,
$
bed+ room would be syllabified as [bsdrum], which is then optionally subject
to affrication.
In addition to such phonological approaches to the syllable, phoneticians
have attempted to provide definitions in terms of its physiological properties.
While each of these has problems associated with it, the syllable has been
defined acoustically in terms of sonority, articulatorily in terms of increasing
and decreasing aperture, and, finally, in terms of motor theory, where each
syllable is seen to correlate with a chest pulse (for discussion and references
see Allen, 1973: 38-45). What is clear is that while the syllable may have
some physical basis, phonological syllable boundaries do not necessarily
correspond to phonetic ones. Hooper (1972: 539), for instance, suggests that
syllabification rules apply "persistently," that is, they reapply at each stage
of a derivation. It is, of course, possible to maintain a phonological syllable
boundary in one place (for example, bed+room may syllabify as /bed$rum/),
but a phonetic one in another place (for example, [be$drum:] = [be$jrum]).
A particularly interesting case of a discrepancy between underlying and
surface syllable boundaries occurs in Maxakali (Gudschinsky, Popovich and
Popovich, 1970). In this language, the following derivations are found:
/CiC/ -+ Ci~C -+ Ciy~C
/CoCf -+ Co; -+ Cow;)C
/CiC/ -+ Ci;)C -+ Ciy~C
/CiC/ -+ Ci~C -+ Ciy~C-+ Ci!J~C
Before certain consonants (especially ft/), a rule of diphthongization converts
underlying monosyllabic /CVC/ to intermediate CVaC. At this point a glide
is inserted as follows: [y] after /i/, [ w] after /o/, and [ y] after fa/ and /i/.
In the fourth line, inserted [ y] is converted to [ lJ] in the context of a nasalized
vowel. Thus, what started out as one phonological syllable is realized
phonetically as two surface syllables. In the spirit of Hooper (1972), we can
propose that resyllabification must take place after glide epenthesis, for
example, Ci<JC becomes [Ci$y~CJ.
6.1.1.1.2 The Syllable in Generative Phonology Despite widespread
192 Suprasegmental Phonology 6.1
use of the concept in the literature, there have been several linguists (and
least one school of linguistics) who have shown a reluctance to accept
syllable as a viable phonological unit (for example, Kohler, 1966).
argument which has been raised against phonological syllables is that,
segments, the location of a syllable boundary within a morpheme can
be phonemic. That is, two morphemes such as ja$pla/ and /ap$laj
differ only in their syllable structure. Of course, we have seen in the bedroom
example in 6.1.1.1.1 that syllable divisions can differ depending on internal
morphological boundaries, but they cannot differ independently of suclt
boundaries. 3 If morphemes could differ only in syllabic structure, then
opposition such as that between bacon [be$kan] and beckon [bt:k$an] (or
$ a vowel is lengthened if followed by a CV sequence (as in a) but not if it
[bt:kan]) could be reinterpreted not as a vowel contrast but as a syllable
contrast, that is, jbe$kanj vs. jbek$;mf. While vowels would in this case be
redundantly tense in open syllables, a problem would arise in distinguishing .
bake [bek] and beck [bt:k], where the only alternative to the 1e! vs. /sf
opposition would be ad hoc syllable distinctions such as /be$k/ vs. /bekf.
One way to prevent such misuse of syllable boundaries is to disallow their
use in phonological descriptions.
Because syllable boundaries can be determined automatically from uni-
versal principles and language-specific facts about the segments contained in
the syllables, genera,tive phonologists have largely worked under the assump-
tion that the syllable is unnecessary in phonology. Instead of writing a rule
of syllable-final devoicing as follows,
C ~ [-voice] 1-$
t
the full segmental determinants of syllable division can be incorporated into
the rule. Thus, in a language where a consonant is syllable-final if it is either
word-final or followed by another consonant, the devoicing rule can be written
with a disjunction:
C ~ [-voice] 1- {#~}
While the use of $ instead of C, V, and # # sometimes simplifies phono-
logical statements (see below), the fact that it can always be avoided is
seen as evidence that it has no phonological status.
Recently, however, arguments have been presented for incorporating the
syllable into generative phonology (for detailed argumentation, see Hooper,
1972, and Vennemann, 1972a). The position of these linguists is summed up
3
The only reservation that need be made is that some words may function as if they have
internal morpheme boundaries. Thus, there is a McAuley Street in Oakland, California,
pronounced [m:Jk$:>1i], which contrasts with McCawley [ma$kh:>li], the name of a famous
linguist (Francine Desmarais, personal communication).
6.1 Suprasegmental Phonology 193
byVennemann (l972a), w.ho states: "All phonological p~ocesses which can be
tated in a general way wtth the use of syllable boundanes can also be sta~ed
. ~ithout them, simply by including the environments of the syllabification
rules in the formula. My contention is ... that in numerous cases such a
formulation would miss the point, would obscure the motivation of the pro-
cess rather than reveal it" (p. 2). (Compare the conclusion of Hoard, 1971 :139-
140.) A single example will suffice.
In the following Modern Icelandic data (taken from Vennemann, 1972a: 3),
a hatur [ha:thY~] 'hatred'
b ofsi [:)f:s1] 'violence'
followed by a CCV sequence (as in b). As seen inc, however,
18
c titra [thi:thra] 'shiver'
certain consonant sequences appear to be exceptional in that they allow the
preceding vowel to be lengthened. The complete set of such sequences
consists of /p, t, k, sf in the first position and jr, j, vf in the second. All other
sequences of two consonants block vowel lengthening.
On this basis, it would be quite complex to present a rule of vowel
lengthening. Vennemann's initial formulation is as follows:
[+stress] ~ [+long] I - C1 (C2 ) V
v
Condition: C2 = r, j, v; if present,
C1 = p, t, k, s
However, the difference between ofsi (where vowel lengthening is blocked)
and titra (where vowel lengthening is permitted) is one of syllabification.
An underlying /VCCV/ sequence will be syllabified either VC$CV or V$CCV
depending on the identity of the consonants involved. Thus, according to the
information just given, ofsi will be syllabified of$si, while titra will be syllabi-
fied as ti$tra. The above rule can now be rewritten to reflect this difference in
syllable structure:
[+stress] -+ [+long] I - $
v
This rule is considerably simpler than the rule involving a condition on
consonant sequences. Of course, in this framework, there would still have to
be statements of where the syllable boundaries occur, and in a language such
as Icelandic, these statements would be quite complex.
6.1.1.2 Other Phonological Suprasegmentals While the trend
appears to be toward general acceptance of the syllable as a phonological
unit, there has been much discussion of whether the syllable may be a unit of
194 Suprasegmental Phonology 6.1
performance rather than a unit of competence (Fromkin, 1968). That
likt; the phoneme, which represents an abstract distinctive unit of sound
is pa:t of th~ speaker's knowled~e of his language, the syllable may simply
a umt reqmred for the productwn or perception of utterances. As
by Fromkin, it may be that phonological units larger than the syllable are
units of performance. The notion of a "breath group," for instance
directly tied to the speech act rather than to an underlying system of '
knowledge. 4 Similarly, Lehiste (1970) argues for sequences of two syllabi~
as a phonological unit: "The disyllabic sequence, consisting of an odd- an~
an even-~umbered syllable, appears as a basic phonological building block
out of whtch words seem to be constructed" (p. 163). In Finnish, for instance
where stress is placed on the initial syllable of a word (and then
in a weaker form on every odd-numbered syllable), Lehiste notes that "the
two syllables comprising the sequences tend to have equal intensity, but each
successive pair has less intensity than the preceding pair" (p. 164). She also
~r~ues that, in Estonian, statements of duration cannot be made with any
mstght except by reference to both the syllable and disyllabic sequences. Of
course, little can be said about where these facts fit into a phonological
system until general agreement is reached on what is considered to be com.
petence (phonological knowledge) and what is performance (use of that
knowledge).
6.1.2 Grammatical Units
It is now generally accepted that grammatical information can often
ex~r~ an influence on the sound system of a language (see 3.3.2). Although
opmwns vary as to what is meant by such entities as morphemes, stems, and
words, phonologists frequently find it necessary to refer to such units in
their analyses. Grammatical units have played an important role in both
the statement of sequential constraints and the statement of phonological
rules. For the moment we shall limit our discussion to the above three units,
whose boundaries are indicated by # # (full word boundary), # (internal
word or stem boundary), and + (morpheme boundary).
6.1.2.1 The Statement of (Underlying) Sequential Constraints Since
the morpheme is defined as the minimal unit of meaning, most linguists
ass~me that morphemes are listed in the lexicon. Thus every item in the
lextcon has a + boundary at each end. While many linguists have accepted
the notion of lexicalized words, that is, polymorphemic forms which for
s:mantic or phonological reasons must be listed in the lexicon (for example,
szlkscreen, as opposed to a silk screen), it is clear that other words may be
generated by means of productive rules of derivational morphology. In a
4
The demarcation of breath ~roups depends, however, on syntactic phrase and clause
boundaries.
6.1 Suprasegmental Phonology 195
proposed word-formation component (Halle, 1973), it is assumed that
input to these rules will be morphemes. Since speakers are claimed to
MVe knowledge of th~ morphemic structure of words (~or example, th: word
ansJormational conststs of the parts trans+ form+ at+ wn + al), the lextcon of
. 1' anguage is viewed as containing an exhaustive list of the existing morphemes.
1
~ecall from 3.3.3 and _3.3.4 that one of the ai~s of pho~olo_gy (especially
enerative phonology) IS to account for alternattons occurnng m allomorphs
~fthe same morpheme (for example, the [ai] of divine as opposed to the [1]
of divinity).
parallel with the question of how the lexicon is structured is the question
of whether sequential constraints should be stated in terms of morphemes.
' Recently, Hooper (1972, 1973) and Vennemann (1972a) have argued for
constraints on syllables (compare Brown, 1970). The need for syllable-struc-
ture constraints may be illustrated by Chomsky and Halle's (1968:417)
examples in a and b:
a blick c abnick (i.e., ab$nick)
b *buick d *agbnick (i.e., *ag$bnick, *agb$nick)
Although the nonsense form blick does not occur in the English lexicon (see
1.6.1), it is well-formed with respect to the phonological properties of English.
On the other hand, the nonsense form bnick is not well-formed, since (it is
claimed) English morphemes do not begin with sequences such as bn-. In
other words, bnick violates a morpheme structure condition (see 4.2.1.2) of
English.
Notice, however, that while the nonsense form in c is well-formed, the
nonsense form in d is not. The reason is that the first can be syllabified as
ab$nick, which yields two well-formed English syllables, but *agbnick cannot
be syllabified in any acceptable way (both *ag$bnick and *agb$nick produce
unacceptable sequences within a syllable). Even if abnick were analyzed as
{a+ bnik/, we would not necessarily expect this to be exceptional in English.
In fact, some phonologists may be tempted to analyze the word agnostic
with a morpheme boundary (that is, a+gnostic), as in a+moral, a+sexual,
etc. (compare the semantically related word gnosis with an initial ortho-
graphic g). What seems relevant, however, is not whether bn or gn begin a
morpheme, but whether they begin a syllable. It may turn out that all se-
quential constraints should be stated in terms of syllables or words, though
much work remains to be done in this area.
6.1.2.2 The Statement of Phonological Rules: Boundaries The syllable
and word are found to be important in the functioning of phonological rules.
In particular, many phonological rules have to be stated with grammatical
boundaries (for example, word-final devoicing), while other phonological
rules cannot assimilate one segment to another segment when certain
boundaries intervene. A number of questions concern the nature of boundaries
196 Suprasegmental Phonology 6.1 6.1 Suprasegmental Phonology 197

in phonology. Where do they come from? How should they be specified clln be expanded to include sequences of segments interspersed with +
example, with features such as [+word boundary])? How many OOlllndlari; tJoundaries, as seen below:
are there in phonology and what is the relationship between them? The
boundaries used by generative phonologists are # # (full word uu,u<"'"''"' A-+ B/C+ _ +D
# (internal word or stem boundary), and + (morpheme boundary). A-+ B/C+ _D
A-+ B/C - +D
addition, a number of linguists (Harms, 1968: 110ff; McCawley, 1968:
Stanley, 1973: 193) have proposed other grammatical boundaries, This claim about the status of + has not been contradicted by any reported
have been represented by symbols such as@,%,&, =, *, !, and -.Some language. .
of these boundaries are language-specific and define the domain of a speciij~, A second function of boundaries is to condition or motivate phonological
phonological process (for example, vowel harmony). Finally, Schane (1973a: rules. That is, there are certain phonological processes which take place only
66) uses the symbol II for a phrase boundary. at a boundary. Some rules take place at a## boundary but not at a# or +
Different boundaries seem to have different strengths, according to the boundary, while other rules apply at both a## and a# boun_da~y but ~ot
following scale: at a + boundary. In fact, in many cases, having a + boundary IS hke havmg
no boundary at all. Recall from 3.4.2 that the rule deleting the fg/ of /ng/
0 + # ## sequences must be made sensitive to a boundary as seen below:
0 1 2 3
g-+0/IJ-#
Of the major boundaries, + is the weakest and # # the strongest. What this
The following derivations are observed:
means is that## has the greatest ability to block a phonological process from
applying across it. One such example, from Fe?fe?-Bamileke, was seen in /brmg ## h~r/ -+ [bnl)~r] (full word boundary)
3.3.2. Another example occurs in Mandarin (see Cheng, 1973: 82-83). In fsmg # ~r/ -+ [sil]:Jr] (internal word boundary)
Mandarin, unaspirated noncontinuants become voiced intervocalically as in {bng + ~r/ -+ [biJg~r] (morpheme boundary)
the following formalization: /fmg~r/ -+ [fil)g~r] (no boundary)

-son]
-cont -+ [+voice] IV_ V
[ -asp
As seen below in a, this rule applies when there is an intervening internal word
boundary (#), but does not apply when there is an intervening full word
boundary, as in b: 5
a /ti # tif -+ [ti di] 'younger brother'
b /liiu ## ti/ -+ [liiu ti] 'old brother' (fig. 'buddy')
Such examples show that## is stronger than #, since it is harder to penetrate.
That # is stronger than + is seen from the fact that the simple morpheme
boundary + is incapable of blocking a phonological rule. Thus, Chomsky
and Halle (1968: 364) have proposed that any phonological rule of the form
A-+BJC_D
5
While some Sinologists may be tempted to view the "neutral" tone as conditioning the
intervocalic voicing, there is good evidence that the neutral tone itself owes its existence
to boundary reduction (see 6.2.1.2.3), which in turn causes the loss of stress (see Cheng,
1973).
In the above forms, the fgf of bring her and singer is deleted, since these
have, respectively, a## and a # boundary. In the forms for longer and
finger, the fgf remains, since neither a + nor the lack of a boundary can
condition deletion of fgf. In other words, it would appear that there are two
internal word boundaries, one of which(#) is like having a word boundar)'
and the other of which (+)is like having no boundary.
The question now is, what role does + have in phonology? Are ther~ rul~~
which are conditioned by +? We see in examples such as fgf-deletwn u
English that + is a weaker boundary than #. This means that it is less
effective both at blocking and at conditioning phonological processes.
There are, however, certain cases where a phonological rule has been con-
ditioned by a + boundary, for example, the formulation of "velar softening"
by Schane (1973a: 95). For our purposes this rule can be formulated as
follows:
k-+s/-+i
This rule is designed to account for alternations such as the following:
electric [k] electricity [s]
critic [k] : criticism [s]
198 Suprasegmental Phonology 6.1 Suprasegmental Phonology 199
6.1

If the forms on the right are analyzed with internal + boundaries (that . d is referred to such works as Chomsky and Halle (1968), Brame
electric+ ity and critic+ ism), then the above rule will not apply to ...(19
rea72era, 1974)' Kisseberth (1972) ' and, for a critique

of Chomsky and Halle,
words as the following: JtOS~ (1972). . ,
The examples which WI.ll n.ow be d~scu~sed are taken fro~ Brame (197~a).
a kill, key, kit, kite (from (kit/? See 3.3.4) In discussing English adjeCtiVes endmg m -atory and -atwe, Brame pomts
b spook#y, hawk# ish, pack#ing
out that there are two stress patterns, as seen below:
The rule will not apply to the forms in a because there is no boundary
.divfnatory b assimilatory
preceding the high front vowel, while it will not apply to the forms in b 1
inflammatory congratulatory
because the boundary present is not the right one. However, notice that what obligatory anticipatory
this means is that the + boundary conditions a phonological rule which is derivative generative
not conditioned by the stronger# boundary. We can conclude either that "'mparative iterative
this is an exception to the hierarchy presented above or that there is something disputative ejaculative
wrong with this (and similar) analyses.
In the words in a, stress is as~i~ed to the syl~able immed~ately .preceding
Since#, but not +, is capable of blocking a phonological process, it is
the -atory or -ative suffix, while m the .words m b, stress IS as~Ign.ed two
not likely that the hierarchy is wrong. While we could simply note this
syllables before the adjective suffix. Smce words. such as derwatwe and
English example as anomalous, there is some reason to consider rewriting
generative have identical syllable structure-bu~ diffihere?t stress :att~~)s-:
the above rule to apply only to specific morphemes. Alternations between one might simply conclude that stress is phonemic (t at ts, unpre 1cta e m
[k] and [s] are limited to lexicalized words (which originally were all
these adjective forms.
borrowed), or to words built on analogy with these words. Thus it appears To do this would, however, miss an important fact abou~ str~ss a.nd word
that the only productive conversion of [k] to [s] is when the word ends in structure in English. Namely, the verbs from which the adJeCtives m a and
~ic. This change takes place before a highly specific set of suffixes (for example,
b are derived are consistently different, as seen below:
~ity, ~ism, -ify, -ize). While this class of suffixes could be abbreviated by the
+ boundary (or by an arbitrary% boundary, if+ were used for something c divine d assimilate
else), it may be just as valid to write the rule as follows: inflame congratulate
oblige anticipate
k -+ sI _ {ity, ism, ify, ize} generate
derive
Or, /k/ in the suffix ic becomes [s] before these suffixes. 6 Although linguists compare iterate
like Stanley (1973) have posited numerous boundaries, it seems likely that dispute ejaculate
further research will provide principled constraints on the use of boundaries The verbs in d end in -ate, while those in c are bisyllabic and do not
in phonology. involve the -ate suffix. What this means is that if we were to try to locate an
6.1.2.3 The Transformational Cycle A major innovation of the internal word boundary in adjectives such as derivative and generative, we
generative school of phonology was the introduction of the transformational would place them differently, that is, derfv#ative and generat#ive. (These
cycle. Receiving its first statement in Chomsky, Halle and Lukoff (1956), it words may involve + boundaries, that is, derfv#at+ive and gener+at#i~e,
receives its fullest treatment in Chomsky and Halle (1968). Since, as we though we have already observed the minor role this boundary plays m
shall see, the application of the cycle depends on boundaries, and since phonology.) Similarly, two words su_ch as divinatory an~ sa!i~atory would
almost all of the examples where its use has been argued involve stress, it is have an internal word boundary at different places, that ts, dwm#atory and
appropriate to consider this issue in the discussion of suprasegmenta1s. salivat#ory. Given the knowledge that adjectives ending i~ -~tory and -ati~e
However, only the general motivation for recognizing the cycle in phonology are derived from verbs, one can predict the stress of the adjective on the basts
will be considered here. For detailed argumentation and exemplification, the of the verb which is contained in it.
There are two good indications that this is in fact what speakers d~, that
6
An alternative approach is to mark those instances of fk/ which undergo velar softening is, they predict the adjective stress on the basis of the verb stress. First: as
with a special diacritic, say [+ VS]; or, alternatively, one could mark those instances of
pointed out by Brame (1972a:68), the word obligatory has two p~ssi?le
/if and /II which condition velar softening with such a diacritic. In either case, the rule of
velar softening would require that forms which undergo it be marked [ + VS]. stress patterns, either obligatory or obligatory. In the first case, the adjective.
 6.1 Suprasegmental Phonology 201
200 Suprasegmental Phonology 6.1
is derived from the verb oblfge; in the second case, it is derived from 111 these examples 1 represents primary stress, 2 secondary stress, and 3
verb obligate. The word sitlivatory cannot be pronounced *salivatory, tertiary stress (see 6.2.1.4). In order to correctly predict these stress patterns,
there is no verb to salive. Since it receives its stress on the basis of the three mechanisms are proposed:
salivate, its stress must be salivatory. A second indication that this is the
(1) rules for stress assignment to lexical (monomorphemic) items
correct way to view English stress is the fact that English speakers are
(2) a compound stress rule
always sure of the stress placement in such adjectives when they cannot (3) a nuclear stress rule
readily locate a verb inside them. A good example is the word pejorative,
which has two pronunciations: pejorative or pejorative. The first stress pattern We shall not discuss the details of (1) here. For our pu~oses a monos~llab~c
is built on the basis of a hypothetical verb to pej6re, while the second is built lexical item receives stress on its syllabic nucleus (that IS, a vowel), while bl-
on the basis of a hypothetical verb to pejorate. Since pejore does not exist, and polysyllabic items receive stress according to other rules. The compound
and since pejorate is not likely to be known, when English speakers see the stress rule assigns stress as follows (1968: 18):
word pejorative written, they are not sure which way to pronounce it. This
analysis seems therefore to have support. 1
The basic principle of the transformational cycle is that a phonological [1 stress]~ [1 stress] 1- .. V ... 1N
rule (usually stress placement) operates on a "word within a word" before v
applying in a second cycle to the complex word as a whole. Rather than If within a noun two vowels have [1 stress] (because the noun is morpho-
representing the internal structure of words by means of the boundaries logically complex, that is, a compound), the first of thes~ rec~ives an additional
# and +, we indicate it by means of labelled bracketing, as follows: [1 stress] specification, while the second by conventwn IS reduced by o~e
[ [deriv]v at+ive ]A [ [gener+at]v ive]A stress level, that is, to [2 stress]. The stress of the compound blackboard IS
In the above bracketing, V stands for verb and A for adjective. The principle thus derived as follows:
of the transformational cycle is stated as follows:
[ [black]A [board]N ]N
Regarding a surface structure as a labeled bracketing [which is generated by
black board ]N (by lexical stress rule)
the syntactic part of the grammar] ... , we assume as a general principle that 1 1
the phonological rules first apply to the maximal strings that contain no brackets, (by compound stress rule)
and that after all relevant rules have applied, the innermost brackets are erased; black board
1 2
the rules then reapply to maximal strings containing no brackets, and again
innermost brackets are erased after this application; and so on, until the maximal
The noun blackboard consists of an adjective black and a noun board, as
domain of phonological processes is reached. (Chomsky and Halle, 1968:15)
indicated by the bracketing in the firstline. In the second line, [1 stress] is
In the above examples, we begin by assigning stress to the innermost brackets, assigned to the vowel of each of these monosyllabic lexical items (which in
that is, [deriv]v and [gener+at]v The stress rules proposed by Chomsky this case are words). At the same time, the innermost brackets are erased.
and Halle correctly assign the stress as indicated. The brackets are then In the third line, the compound stress rule has assigned [1 stress] to.bla~k,
erased, and we obtain [derivative]A and [generative]A, with no further while the [1 stress] of board is automatically reduced to [2 stress], y1eldmg
modifications needed. the correct stress pattern. . .
In order to show how a stress assignment rule may apply in a cyclical The nuclear stress rule applies in just those cases where two lexical Items
fashion, let us turn to a different kind of stress phenomenon. Chomsky and . (words) come together in a phrase but are not compounded. It can be
Halle (1968:20--22) point out that the same morphemes, black, board, and formalized for our purposes as follows:
eraser, combine to yield three different stress patterns:
1
black board-eraser (board eraser which is black)
2 1 3 [1 stress] ~ [1 stress] I V ... - ... 1NP
blackboard-eraser (eraser for a blackboard) v
1 3 2
black board eraser (eraser of a board which is black) If within a noun phrase two vowels have [1 stress], the second of these
3 1 2 receives an additional [1 stress] specification. Again, by convention, the
202 Suprasegmental Phonology 6.1 6.2 Suprasegmental Phonology 203
[1 stress] in first position is reduced by one stress level. The noun .In each of these derivations, stress is first assigned within the innermost brackets
black board is thus derived as follows: (that is, to units which do not have internal brackets); these brackets are
[ [black ]A [board lN ]NP then erased and stress is assigned within the remaining innermost brackets,
[ black board lNP (by lexical stress rule) and so on. Derivation b best illustrates the principle of the transformational
1 1 cycle. In the first cycle, [1 stress] is assigned by the lexical stress rule to
black board (by nuclear stress rule) (bhick]A, [bmird]N, and [ eniser]N, as indicated. Mter lexical stress has been
2 1 assigned, we are left with the bracketing [[black board]N eraser]N. Thus,
As in the pre:io~s derivation,. w_e begin with the words [black]A and [board)N. looking at the innermost bracketing, we see that the compound stress rule
However, this time they are JOined together not as a compound noun but will apply to [black board]N, changing its 1-1 stress pattern to 1-2. Mter the
a noun phrase, that is, an adjective modifying a noun. By the lexical .a$ brackets around [black board]N are erased, we are left with [black board
rule, [1 stress] is assigned to the vowel of each of these words. The .......,.utosT eraserJN At this point a second application (or cycle) of the compound stress
brackets are then erased. At this point the compound stress rule cannot rule applies, converting the input 1-2-1 stress to 1-3-2. That is, (1 stress]
since black and board are joined not as a noun but as a noun phrase. is assigned to the leftmost member of the noun compound, black, thereby
nuclear stress rule then applies, assigning [1 stressJ to board and requiring that all other stresses be reduced by one. By use of the .trans-
the [1 stress] of black to a [2 stress] specification. formational cycle, therefore, the complex stress patterns of English can
With these rules we are able to account for the stress differences between be adequately accounted for. 7
l 2 2 l
blackboard and black board. At this point we are ready to move on to the
three stress possibilities which are observed when black, board, and eraser
are combined. The derivations are given below (Chomsky and Halle, 1968:21):
a 'board eraser which is black' 6.2 Suprasegmentals of Prominence
[ [black]A [ [board]N [eraser]N 1N ]NP
The word prominence is used as a cover term to include stress, tone,
[ black [ board eraser lN ]NP (by lexical stress rule)
1 1 1 and duration (see Voorhoeve, 1973:1n). While to some linguists only these
[ black board eraser ]NP (by compound stress rule) features are true suprasegmentals, other linguists have analyzed vowel har-
1 1 2 mony and nasalization suprasegmentally as well (see 6.3). The features of
black board eraser (by nuclear stress rule) stress (intensity), tone (pitch), and duration (length) are always present in all
2 1 3 utterances (Martinet, 1960: 75). Thus, any utterance in any language is
b 'eraser for a blackboard' characterized by differing degrees of loudness, melody, and rhythm. In
[ [ [black]A [board]N lN [eraser]N ]N addition, it is noted that these three aspects of the speech signal, corresponding
[ [ black board lN eraser )N (by lexical stress rule) respectively to stress, tone, and duration, can, unlike segmental features, be
1 1 1 isolated and extracted as a pattern on an utterance. As such, each of these can
[ black board eraser )N (by compound stress rule) be easily demonstrated in a medium other than speech. On a guitar,for instance,
1 2 1 differing degrees of loudness depend in large part on the force with which
black board eraser (by compound stress rule) a string is plucked, different melodies are obtained by plucking different
1 3 2
notes, and different rhythms are obtained by varying the duration of each
c 'eraser of a board which is black'
pluck. Unlike voicing, nasalization, affrication, etc., stress, tone, and duration
[ [ [black]A (board]N JNP [eraser]N ]N
are "overlaid functions" on segments (Lebiste, 1970: 2) which can be produced
[ [ black board lNP eraser ]N (by lexical stress rule) independently of these segments both by the human voice (for example,
1 1 1
[ black board bumming) and by other modes of production.
eraser lN (by nuclear stress rule)
2 1 1 1
black board Notice, however, that the stress patterns can be modified by means of contrastive or
eraser (by compound stress rule) emphatic stress. Thus, if one contrasts a bldckboard eraser with a bldckboard stand, eraser
3 1 2
may receive contrastive stress (e.g., "I said blackboard eraser, not blackboard stand).
 204 Suprasegmental Phonology 6.2
6.2.1 Stress
Of the three suprasegmentals of prominence, stress receives by
the most developed treatment in the literature. While this can be '"ro'"""
attributed to the fact that most European languages are stress
recent intensive work on tone languages (see 6.2.2) provides a pers_recltive~J
in placing stress within the wider context of prominence.
6.2.1.1 What Is a Stress Language? Stress has been defined
ba~ically two ways: first, in terms of its phonetic properties; second, in term8 .
of Its linguistic function. We shall first emphasize the function of (word) stre
and then turn to its phonetic realization (compare the approach of Garde 1 ss
In looking at various languages of the world which are said to be ~haraC.,
terized by stress, it is quickly observed that stress has a culminative function.
The purpose of assigning stress, as in the first syllable of data and the second. .
syllable of deter, is to mark one syllable per word as carrying prominence
That is, there is a culmination of prominence on one syllable, and only on;
syllable per word (or stress unit) can receive this prominence. While in all
stress languages prominence is culminative, it is at this point that stress
languages begin to differ.
The major distinction that must be drawn is between free vs. fixed stress.
In a language with free stress, prominence can occur on different syllables
(for example, first, last), depending on the word. Thus, in Russian the two
words muka 'torture' and muka 'flour' are distinguished by the fact that in
muka stress is on the first syllable while in muka it is on the second (Trubetz-
koy, 1939: 188). That we are dealing with a stress language is seen by the fact
that there are no Russian words pronounced muka or mukii. Since stress is
~ulminative, there can be no word where all syllables are marked by prom-
I~ence, nor can there be a word where no syllable is marked for prominence. s
Smce stress can occur on the first syllable in one word but on the second in
another, stress is said to be phonemic in these languages.
.L~nguages which restrict the placement of stress to one particular syllable
Wit~m each word are said to have .fixed or nonphonemic stress. Thus, stress is
assigned to the first syllable in Hungarian, to the last syllable in Turkish, and
to the. penultimate (sec~nd from end) syllable in Polish. In these languages
stress IS completely predictable. In a language such as Russian, stress will often
have to be marked on lexical items; stress thereforeacquiresa/exicalfunction.
In a language such as Hungarian, where the first syllable of every word is
stressed, ~exical items need not be marked for prominence. Instead, a rule of
stress assignment figures among the phonological rules of the language:
$--+ [+stress] f ## _
8
There are apparently languages which have words lacking stress, for example, Seneca
(Wa~lac~ Chafe, personal communication), Hungarian (Robert Hetzron, personal com-
mumcation), and others.
6.2 Suprasegmental Phonology 205
A syllable is stressed in word-initial position. 9 In Hun~arian an_d other
tanguages with fixed stress, stress may have a demarcatwe functwn (see
Martinet, 1960: 87). That is, stress signals a word boundary. In this respect
it can be compared to languages such as German and Arabic where a glottal
stop is inserted stem-initially before a vowel (that is, before the# boundary).
Given a stress in Hungarian, we know that we are at the beginning of a word.
Thus, it would appear that stress is an aid in processing utterances. In a
language with demarcative stress, each stress tells us where we are in the
word. In a language with lexical stress, we can merely correlate stress with a
different word, though we do not know exactly where in the word we are.
6.2.1.2 Factors Determining Stress Placement In languages with
lexical stress, the placement of the stress within a word is part of the under-
lying phonological form. Therefore, no rules of stress assignment are needed.
In languages with predictable stress, prominence is assigned according to
grammatical and sometimes also phonological factors.
6.2.1.2.1 Grammatical Factors The most obvious grammatical
factor in determining stress placement is the word boundary(##). As noted,
some languages assign stress to the first syllable, others to the penultimate
syllable of each word, etc. The grammatical boundary which is relevant
for stress placement may vary somewhat from language to language. Thus,
in French, stress is placed on the last syllable of each sense group, for example,
de Ia mairie 'from the town-hall,' Ia Tour Ei.ffel 'the Eiffel tower.' If each
word were to receive stress on its last syllable we would have the incorrect
*Ia Tour Ei.ffel. In French, then, the word is not a relevant category for stress
placement. In other languages, stress is automatically placed on the stem of
each word. In this case the relevant boundary is #, instead of the full word
boundary # #.
In addition to boundary information, stress rules must sometimes make
reference to grammatical categories. In 3.3.2 it was observed that in some hi-
syllabic noun-verb pairs in English, stress is assigned to the first syllable in
nouns (for example, convert), but to the second syllable in verbs (for example,
convert). 10 In Spanish, stress is assigned to the last syllable of infinitives
(for example, decir 'to say', not *decir). Since stress is expected on the
penultimate syllable (subject to syllable weight; see 6.2.1.2.2), infinitives
constitute an exception to the general rule. One possibility is to have a
morphologized stress assignment rule which would make explicit reference to
the category "infinitive," much as the English rule must refer to the categories
9
In generative studies such rules are usually written as
V-+ [+stress]
since stress is seen to be a property of syllabic segments.
10
There are, however, important exceptions to both of these patterns, some systematic,
some idiosyncratic, e.g., to revel vs. to rebel.
 206 Suprasegmental Phonology 6.2
"noun" and "verb." On the other hand, J. Harris (1969:177ff) proposes
infinitives have an abstract final /e/, that is, decire 'to say.' In this case
can first have penultimate stress assignment (decire) and then final.
deletion (decir). The form decfr would in this case be only a special t:x1::en'ti""~
to the general pattern of stress assignment in Spanish. Since hi~~ : __ ,.
there was such a vowel on infinitives, we at least gain some insight into
a morphologized rule (that is, final stress assignment in infinitives) can
into being (see 5.2.6.2).
6.2.1.2.2 Phonological Factors While languages with fixed
single out a particular syllable of a word for stress, that syllable may be more
or less "stressable" depending on its phonological structure. Recall from
6.1.1.1.1 that the syllable was divided into an onset and a core. In a
syllable, C is the onset and VC the core. In many languages, a syllable
core consists solely of a short vowel (V) cannot be stressed and stress must
pass to a neighboring syllable. Such a syllable is said to be light. A syllable
whose core consists of a long vowel (V :), a VV or VC sequence, or com~
binations of these, can be stressed and is said to be heavy. This distinction in
syllable weight is therefore an important phonological variable in the state,.
ment of stress placement (Newman, 1972; Allen, 1973).
The best known example of syllable weight comes from Latin. As seen in
the following examples (Allen, 1973: 51),
a refe:cit
b refectus
c reficit
stress is assigned to the penultimate syllable in a and b, but to the ante~
penultimate syllable in c. This difference is, of course, conditioned by
syllable weight. Stress is normally assigned to the penultimate syllable in
Latin, except when that syllable is light. In this case, the stress is assigned to
the antepenultimate position, as in reficit, where the penultimate syllable fi
is light.
In many languages stress can be assigned only to a heavy syllable. Thus,
Jakobson (193Ia:l17) reports that Classical Arabic assigns stress to the
first heavy syllable of a word. One important observation is that all languages
with a heavy vs. light syllable dichotomy have a vowel-length contrast, that
is, CV contrasts with CV:, which patterns with CVC. If this were not the
case, we would simply have a contrast between open (CV) and closed (CVC)
syllables. Apparently no language requires that stress be assigned only to
closed syllables. Thus, in the absence of CV:, a CV syllable will always be
able to accept stress. Since the same CV functions as a light syllable in
languages with a vowel-length contrast and as a syllable equal in weight to
eve in languages without a vowel-length contrast, the explanation for
6.2 Suprasegmental PhotWlogy 207
weight as a factor in stress placement cannot be purely phonetic.
6.2.1.3, it will be shown that stress has a tendency to lengthen vowels.
.'fh~s, if Latin reficit were to receive stress on its penultimate syllable, the
vowel fi/ of this syllable would tend to lengthen, and /fi/ would threaten to
lilerge with /fi:/. In order to avoid this merger, stress is shifted, hopefully to
nnd a heavy stressable syllable. In some languages (for example, Eastern
Chemeris [Itkonen, as reported in Kiparsky, 1972:190]) stress is retracted
back further and further until it finds a compatible syllable. In the event that
there is no preceding heavy syllable, stress is expected on the first syllable
(for example, Latin reficit, where re is a light syllable).
6.2.1.2.3 Factors Determined by Stress Placement In the foregomg
discussion we have emphasized the linguistic function of stress. Although
stress is seen to be a grammatical feature (which can become part of a lexical
entry), somehow speakers have to provide phonetic cues so that the stress
can be identified by listeners. Since stress is culminative, it can be assumed that
the intention of the speaker is to give prominen~e or saliency to the stressed
syllable. While it was long believed that the primary phonetic cue of stress
was intensity (that is, the energy expended in producing it), phonetic investi-
gations have revealed that intensity is not a reliable correlate of stress (Mol
and Uhlenbeck, 1956; Fry, 1955, 1958). Instead, pitch and duration (in that
order) are much more effective cues of stress than intensity. This has led
some scholars (for example, Bolinger, 1958:111) to conclude that "pitch is
our main cue to stress."
Part of the reason that stress was viewed in terms of intensity was the
feeling that it would have to be radically different from tone (see 6.2.2).
However, since pitch is seen to be the most important phonetic signal of
stress, and since pitch is clearly the most important cue of tone, the difference
between stress and tone is a linguistic one and not a phonetic one. This
explains why placing a stress on a given syllable can cause modifications of
the segments over which it has domain. Correlating with stress is a changing
pitch (usually rising from an unstressed to a stressed syllable and falling from
a stressed to an unstressed syllable), greater duration (for example, vowel
lengthening in a stressed open syllable), and greater force of articulation (for
example, the tendency for consonants to become aspirated or geminated).
While the pitch characteristics of a word such as data, with stress on the
first syllable, are not perceptibly different from a sequence of high followed by
low tone in an African tone language, linguistic tone has not been shown to
have any of the above effects: Since both stress and high tone correlate with
prominent pitch, it must be concluded that the segmental effects of stress are
due entirely to its culminative function. Both vowel lengthening and consonant
fortition signal the prominence of a syllable which has culminative stress.
Since stress has these intrinsic properties associated with it, it is not
surprising to find languages phonologizing (see 5.2.5) these properties into

208 Suprasegmental Plwnology 6.2
rules of the language. Numerous cases of strengthening in stressed
and weakening in unstressed syllables are attested, some of which,
example, consonant fortition in Finnish, were discussed in 5.2.4.2. In
tense stressed mid-vowels undergo lengthening and then diphthongization
follows:
pede --+ pe:de --+ piede 'foot'
b6no --+ b6:no --+ buono 'good'
Stress causes vowel lengthening and long vowels tend to diphthongize
raise or both (see Labov, Yaeger and Steiner, 1972). An interesting case
weakening in unstressed syllables is reported for Mandarin Chinese
1973) In the following derivation,
fli pa/ --+ [If oo J 'fence'
the second syllable is unstressed. Three things happen as a result: (1) the
low back vowel fa! is reduced to schwa, (2) the voiceless stop fp/ is weakened
to [b], and (3) the high tone of fpaf is reduced to "neutral" tone, which in
the above example has low pitch. Since these three adjustments are all
associated with lack of stress, Cheng (1973: 83) concludes: "All the segments
in a neutral-tone syllable become lax."
6.2.1.3 Natural Stress Rules Given that the function of stress is
to highlight a particular syllable of a word, any rule which contributes to the
identification of that syllable's prominence will be considered natural; by
the same token, any rule which detracts from the prominence of that syllable
will be considered less natural (or, conceivably, unnatural). Rules of stress
placement can be evaluated for both their conceptual and their phonetic
naturalness.
6.2.1.3.1 Conceptual Naturalness Conceptually, since stress ideally
demarcates word boundaries, the more regular the stress assignment, the
more successful it is in fulfilling its linguistic function. A stress rule which
requires morphological information (that is, class categories) or which refers
to syllable weight is less natural, conceptually, than a rule which operates
across the board. In addition, a rule which places stress closer to a word
boundary is more natural than a rule which places stress further from a
word boundary, at least from a conceptual point of view. In other words,
stress tends to stay close to the beginning or end of a word. We therefore do
not expect to find many languages which stress the third syllable of a word,
since this would require much more calculation on the part of the speaker
and the hearer than would languages which place stress on the first syllable.
That syllable weight adds to the conceptual complexity of a stress rule is seen
from the fact that the Latin phrase b6nacalfgula allows two possible seg-
mentations (Martinet, 1960: 87):
a b6na caligula
b b6naca ligula
6.2 Suprasegmental Phonology 209
b6na receives stress on its penultimate syllable, while calfgula receives
:;;son its antepenultimate syllable (its penultimate syll~ble gu is.~ight and
therefore cannot accept stress). In~ both. b6naca ~nd bgula. receive stress
on their antepenultimate syllables, smce their respective penultlmat~ syllables
na and gu are light and cannot take stress. As pointe? ?ut by ~artmet, there
is no way to predict that b6nacaligula should be dlVlded up mto words as
a rather than as in b. If stress were completely regular, however, a
:ould be pronounced b6na caligula and b bonaca ligula. . .
6.2.1.3.2 Phonetic Naturalness While conceptual considerations
ould tend to have stress realized either word-initially or word-finally,
;enultimate position is favored over final position by languages. ~~ere ap-
pears to be an asymmetry, since the two most highly favored positions for
stress are the first syllable of a word and the second syll~ble from. ~he
end of a word. The attraction of stress from final to penultn~ate positi~n
can be explained by recourse to phonetic naturalness. As pomt~d ou~ m
6.2.1.2.3, Bolinger (1958) and others have established tha~ (chan~mg) pitch
is the primary acoustic cue of stress. Consider the approximate pitch values
in the following English words:
perfect (verb) [_ ") 1
perfect (adj.) r- _1
[_ - _ ]
perfection
These words exhibit stress in initial, final, and medial position, respectiv~ly.
While the stressed syllable in perfection rises in pitch from the precedm~
unstressed syllable and falls in pitch to the following unstr~ssed .syllable,. It
appears that the fall is perceptually more salient than the :1se. ~lfs~, notice
that since perfect goes from a high pitch to a low pitch, there IS no nse m~olved
at all. Second, if the rise from an unstressed to a stressed syllable wer~ pn~ary,
there would be no totally satisfactory way to explain the fall which IS ob-
served in perfect. Since a low-high sequence in a tone language does not
involve such a final fall, this fall cannot be attributed entirely to th~ fact t~at
this stress is in utterance-final position. In fact, if the final stress m perfect
did not fall, that is, if the pitch pattern were L -], linguists would probably
be inclined to call English a pitch-accent or tone-placement language (see
6.2.3.1). Since a monosyllable such as b6y or gi~l is realiz.ed as. [ ")] rather
than as *[ J] in stress languages, we can assoctate a falhng pitch contour
with underlying stress. . .. .
Accepting this position, we can now say that penultlm~te positiOn iS
favored over final position, because a falling contour reahzed over two
syllables requires less articulatory effort (note the tende~cy to level out
contour tones in tone languages-see 6.2.2.3.1.2) and mvolves greater
perceptual prominence-that is, the high pitch of perfect is more salient th~n
the falling pitch of perfect (note the tendency of falling tones to become illld
210 Suprasegmental Phonology 6.2
or low in tone languages). Since language is characterized by downglide
utterance-final position (see footnote 16), a high pitch on the peJrrultnrtatck
syllable followed by a low pitch on the final syllable maximizes the fall
is thus favored over realizing the falling contour on the one final syllable.
Thus, a rule which assigns penultimate stress is more natural from a phonetic,
point of view than a rule which assigns final stress.
That initial and penultimate positions are the most natural for stress
placement is dramatically confirmed in Auca. In this language, words
consist of two parts, the stem and the suffix complex, both of which can be
polysyllabic. As reported by Pike (1964:186-187), there are two primary
stress rules or "wave trains." First, counting from the end of the word,
suffixal syllables receive "alternating" stress on every even-numbered syllable.
Thus, the penultimate syllable will be stressed, as well as the ante-ante-
penultimate syllable, etc. A second rule of alternating stress assigns stress to
every odd-numbered syllable counting from the beginning of the word. Thus,
the first, third, etc. syllables of a stem will be stressed. A word with four
stem syllables and four suffixal syllables will therefore be stressed
CVCVCVCV#CVCVCVCV. In this case a perfect stressed-unstressed pattern
is obtained, though Pike points out that interesting complications arise
when two stresses "bump" at the stem boundary.
The Auca example reveals that stress rules can apply iteratively on every
other syllable, starting from the syllable receiving primary stress. Similarly,
Lehiste (1970:163-164) points out that Finnish receives initial stress and
then weaker alternating stresses on each odd-numbered syllable, that is,
c\Tcvcvcv... It appears that alternating stress facilitates the processing
of stress. This is particularly clear in a language which has penultimate stress,
where alternating stresses establish a rhythm which crescendoes in penultimate
position.
6.2.1.4 Degrees of Stress In discussing the culminative nature of
stress, it was stated that there can be only one (primary) stress per word.
As seen in the discussion of the transformational cycle (6.1.2.3), however,
we spoke of three levels or degrees of stress, which were indicated by the
integers 1, 2, and 3 (that is, primary, secondary, and tertiary stress). It is
sometimes claimed that English recognizes four levels of stress (Trager and
Smith, 1951; Chomsky and Halle, 1968). It is difficult to support this claim,
since stress is presumed to be a grammatical (mental) feature, and not
directly a phonetic one. In other words, when one stress is judged to be more
prominent than another, this decision may be made on the basis of the
grammatical nature of an utterance rather than its phonetic nature. In the
English utterances fireman [fairm;}n] and fire man [fair mren], the first
Oexicalized) form is judged to have 1-0 stress while the second has 1-2 stress.
We know that the second syllable of fireman receives less prominence
phonologically than the second syllable of fire man, since its vowel is reduced
6.2 Suprasegmental Phonology 211
While this reduced vowel can be expected to be shorter in duration
the nonreduced vowel [re], the two words appear to have the same
pitch characteristics. As pointed out by Lehiste (1970:150), it may simply be
that a syllable receiving nonprimary stress may be heard as stressed because
at some underlying level a major stress is assigned to this syllable (see Chomsky
and Halle, 1968:26n). In Lehiste's words, "we 'hear' the underlying phono-
logical form." The words/ire and man receive [1 stress] by the lexical stress
rule referred to in 6.1.2.3. In fire man the [I stress] of man is reduced to
(2 stress] by the compound stress rule. In fireman, an additional application
of the compound stress rule may apply, since this form is lexicalized (that is,
it is learned as a single form rather than created by a productive rule). The
vowel therefore reduces to schwa, since the man syllable is felt to be less and
less related to the individual word man, which receives [1 stress]. Thus,
speakers may feel that a syllable has greater or lesser stress according to their
ability to relate this syllable to another occurrence where it has [1 stress].
In English, the less able speakers are to see such a relationship, the more
likely the vowel of such a syllable will be reduced to schwa.
What this means is that speakers may rate syllables on the basis of their
potential ability to be stressed. This may mean the possibility that a morpheme
may be unstressed in one word but the same morpheme may be stressed in
another word; or it may reflect that syllable's potential for receiving emphatic
or contrastive stress. Thus, the only way to emphasize fireman (for example,
'I said fireman, not yeoman') is by placing greater stress on fire. On the other
hand, stress can be shifted to the syllable man of fire man, as in the sentence
'I said fire man, not fire woman' (compare apple pie vs. apple pie). Thus, the
man of fire man may be viewed to be more stressed than the man of fireman
because it can receive contrastive stress. Of course, it can receive contrastive
stress because it retains its literal meaning 'man' as opposed to the meaning
'person' infireman.
Such demonstrations have caused scholars such as Lehiste (1970:150)
to conclude (compare Weinreich, 1954 for Yiddish): "It appears probable
that word-level stress is in a very real sense an abstract quality: a potential for
being stressed. Word-level stress is the capacity of a syllable within a word
to receive sentence stress when the word is realized as part of the. sentence."
For this reason, secondary stresses often have only "remnants'' of primary
stress characteristics. While they normally lack the pitch correlates of
primary stress, they may have other segmental correlates (for example,
failure of a vowel to reduce to schwa in English). A particularly clear example
of this is presented from Spanish by Brame (1974). Brame points out that in
Spanish, under certain conditions, when the theme vowel fe/ of the third
conjugation of verbs is stressed, it becomes [i]. Thus, fdebes/ you owe' is
pronounced [de~es], while fdebedo/ 'owed' (past participle) is pronounced
[de~i5o]. We do not obtain *[di~es], because the first fe/ is not the theme
212 Suprasegmental Phonology 6.2 6.2 Suprasegmental Phonology 213
vowel, and we do not obtain *[de pis], because the theme vowel is 6.2.2.1 What Is a Tone Language? Pike (1948:3) defines as tonal
stressed. The stressed vowel, of course, receives [I stress]. anY language "having significant, contrastive, but relative pitch on each
An interesting problem arises in the derivation of adverbs in syllable." As seen in the following examples (George, I970:102),
The relevant adverbs are formed by suffixing mente to the past n~t't,,..;;_,
high [ba] 'to be sour'
of a verb, for example, resign~da 'resigned,' resign~dam~nte tnid : (ba] 'to cut'
As just indicated, the past participle would by itself receive [I tow [ba] 'to count'
its penultimate syllable. In the adverb, however, primary stress is ass:llmen\:
Nupe has a phonemic contrast between high tone, mid tone, and low tone on
to the mente suffix, and the underlying primary stress of the past n<>,,t,,.,.;_,,,<
any given syllable. The tone marks used in this section are as follows:
is reduced to [2 stress]. Since the correct form involving a verb of the
conjugation is debidamente 'justly,' that is, with raising of stressed fef to For African languages:
the following derivation is proposed: high /a/= H rising tat = R
tnid /li/ = M falling /a/ = F
[ [debeda] mente ] low /a/= L downstep f'aj = 'H
/'a/= 'M
[ [debeda] mente ] by penultimate stress rule For Chinese (Mandarin)
1
tone 1 /li/ = rJ (high)
debida mente ] by vowel raising rule tone 2 /a/ = [._;] (high-rising)
1 tone 3 /a/ = [ v] (dipping/falling-rising)
debida mente by penultimate stress rule tone 4 /a/ = [ 1 ] (high-falling)
2 1
Unlike stress, different tones can lexically contrast in a given phonological
In the first cycle, stress is assigned penultimately to jdebeda/ and then the environment. In a stress language it suffices to state where in the word (that
stressed fe/ is raised to [i]. 11 In the second cycle, penultimate stress is assigned is, on which syllable) primary stress is placed. Thus in a bisyllabic word there
to /mente/ and the primary stress on /debeda/ is reduced to [2 stress], are two possible patterns: stressed-unstressed or unstressed-stressed. In a
Although this [2 stress] does not necessarily carry with it the pitch chara~ tone language such as lgbo (see Welmers, 1970), as seen below,
teristics of a primary stress, speakers will recognize that the syllable hi high-high [akwa] 'crying'
receives [I stress] in the word debida 'owed,' and that in order for the high-low [akwa] 'cloth'
underlying fe/ to become [i] there must be some stress associated with it low-high [akwa] 'egg'
2
Thus, mentally bi receives more prominence than either the preceding de low-low [akwa] 'bed'
or the following da syllable (see Hooper, I973 for a noncyclic approach four possible tone patterns are found, since H or L can occur on either
Spanish phonology). syllable. In tone languages, there are sometimes restrictions on the occurrence
of tones, which can be either phonological (for example, the last tone of an
utterance must be L) or grammatical (for example, the noun-class prefixes of
6.2.2 Tone Bantu have L tone).
While stress was said to be of a culminative nature, having a Since these restrictions can sometimes be quite pervasive, this means that
demarcative function in many languages, tone more directly resembles there will be a lot of redundancy in the distribution of, say, H and L tone.
segmental phenomena. Although most of the discussion of this section will For this reason, Welmers (1959:2) suggests that Pike's definition of one tone
focus around African tone languages, tone is found in most parts of the per syllable is too strong. Instead, he proposes_ that "a tone language is a
world (for example, Southeast Asia, Australia, Mexico). language in which both pitch phonemes and segmental phonemes enter into
the composition of at least some morphemes." Thus Nupe /ba/ 'to be sour'
11
Brame puts the vowel-raising rule after the last cycle, although the correct output is ob- consists of the segmental phonemes /b/ and fa/ and the pitch phoneme/'/.
tained either way. By placing it within the first cycle, however, the hypothesis can be As we shall see, some morphemes (for example, grammatical affixes, pro-
advanced that only primary stress has the typical segmental effects of lengthening, raising, nouns) may lack a pitch phoneme (tone), while other such morphemes may
diphthongization, etc. consist solely of a tone (with no segmentals).
114 Suprasegmental Phonology 6.2
Pike (1948: 5) draws a distinction between register tone languages
contour tone languages. In a pure register tone language, tonal
consist of different levels of steady pitch heights, that is, perceptually,
tones neither rise nor fall in their production. The Nupe and Igbo exalffiJ:lles
given above are of this type. A pure contour tone language consists of
tones which are not level in their production but rather rise, fall, or rise
and fall in pitch. In general, African tone languages are of the first type While
Oriental languages are of the second. Thus, the following minimal pairs'
reveal that in Peking Mandarin, three of the four tones are contours:
/rna/ 'mother' L]
/rna/ 'hemp' [-']
/rna/ 'horse' [-...;]
/rna/ 'scold' [I ]
Since the tone on 'mother' is a level H tone, Peking Mandarin is not a pure
or consistent contour tone language.
While in all of the above examples tone has been seen to exhibit a lexical
function, in many if not most tone languages tone also has a grammatical
function. Thus, in Shona, tone is used to distinguish between a main and a
relative clause:
mwana akawuya 'the child came'
mwana akawuya 'the child who came'
In other languages, tone serves to mark different verb tenses, possession and
even negation.
6.2.2.2 The Lexical Representation of Tone Current interest in tone~
research can be divided into two general categories: (1) the lexical repre.
sentation of tone and (2) the nature of tone rules. A number of debates have
centered around the first of these.
6.2.2.2.1 Segmental vs. Suprasegmental Representation of Tone
Perhaps the most lively debate in generative studies of tone centers around
the issue of whether tone is a segmental or a suprasegmental phenomenon.
Given the tonal contrasts of /ba/, fbaf, and fba/ in Nupe, the question is
whether H, M, and L should be features assigned segmentally to the vowel
fa/ or whether tone should be assigned to units larger than syllabic segments
(vowels, syllabic nasals and liquids, etc.). Although this controversy is
usually stated as a debate between suprasegmentalists and segmentalists,
there are at least two separate questions to be answered. The first is whether
the syllable is a viable unit for tonal representation and the statement of tone
rules. Although tone is sometimes maintained to be a feature on syllables
(Wang, 1967:95), generative phonologists have, for reasons which we have
seen, tried to avoid syllables and speak instead of syllabic segments as carrying
tone. Thus, instead of saying that H tone is assigned to the entire syllable
fbaf 'to be sour' in Nupe, the underlying H tone is assigned to the [+syllabic]
6.2 Suprasegmental Phonology 115
.segtnent /a/. It appears that the syllable approach and the segment approach
are readily translatable into each other. That is, whether w~ say that the H
tone of [ba] is assigned to the syllable /ba/ or simply to the vowel fa/, the
same insights into the tonal structure ofNupe are obtained. We can assume
that this is due to the fact that syllables are defined in terms of segments and,
as a result, it is always possible to avoid talking about syllables and talk
instead of the segments which define them (see 6.1.1.1.2). Thus, tone is fre-
quently assumed to be a segmental property (Schachter and Fromkin, 1968;
Woo, 1969; Maddieson, 1971).
A totally different debate centers around the question of whether tone can
ever be assigned to underlying grammatical units, for example, morphemes
or words. In both the segment and the syllable approach, tone is assigned to
an underlying phonological unit, while in this case the possibility of assigning
tone to a grammatical unit is considered. In the Igbo examples given earlier,
a phonological approach would assign H or L to each syllable or syllabic unit,
while a grammatical approach would assign H, L, F, orR to each word (or,
conceivably, to each morpheme). In the case ofF and R, a falling tone would
be realized over two syllables as a H followed by a L, while a rising tone
would be realized as a L followed by a H.
Depending on whether underlying tone is assigned to a phonological unit
(either the syllable or syllabic unit) or a grammatical unit (morphemes,
words, or perhaps tone phrases), considerably different tone systems result.
Let us say, for instance, that it has been established that a certain language
distinguishes H and L tone on monosyllabic words. When we look at words
which are longer than one syllable, we expect one of two situations. First, if
tone is assigned to a phonological unit, we expect four tone patterns on
bisyllabic words (H-H, H-L, L-H, L-L), as seen abovefor Igbo. If, on the
other hand, tone is assigned to words, only two tone patterns are expected
on bisyllabic words (H-H and L-L). That is, a whole word would take either
one tone (H) or the other (L), and this tone would be realized throughout the
word, no matter how many syllables that word is composed of. As argued
by Leben (197lb; 1973a,b), a language having only two tonal possibilities
independent of the number of syllables in a word would remain unaccounted
for if tone were not assigned to underlying grammatical units. Leben thus
states (1971 b) :
One fact about Mende [compare Dwyer, 1971] which points to the appro..
priateness of suprasegmental representation is that a constraint must be stated to
rule out the sequence HLH on all morphemes. The following sequences, for
example, are impermissible: *CV; *CVCV; *CVCVCV. If the sequence of tones
is represented as a feature on the morpheme [or word], a single statement of the
constraint will cover all morphemes regardless of the number of their syllables;
such a general statement might not be formulable if we took tone as a segmental
feature. (p. 197)
 6.2 Suprasegmental Phonology 217
216 Suprasegmental Phonology 6.2

Leben also shows that certain tone rules cannot be properly understood in . pounded types of a register system. (3) The beginning and ending points of the
glides [contours] of a contour system cannot be equated with level tonemes in
segmental framework, since whole (bisyllabic) morphemes are raised
the system, whereas all glides of a register system are to be interpreted phonemic-
lowered in pitch as a unit. allY in terms of their end points. (4) In the printed material examined contour
If Leben is correct in representing underlying tone suprasegmentally itf systems had only one toneme per syllable, whereas some of the register tone
some languages (for example, Mende, Maninka, Hausa), then Pike's de- languages, like the Mazateco, may have two or more tonemes per syllable. (p. 8)
finition of a tone language as having contrastive pitch on each syllable (see
6.2.2.1) must be modified or abandoned (compare Welmers' definition). In We have already mentioned point l. Thus Mandarin Chinese is considered
a recent study of Tamang phonology, Mazaudon (1973:85-92) presentsa' . - to be a contour tone language, while Nupe and Igbo are register tone lan-
number of arguments for recognizing word-tone in this and presumably other . guages. However, it is not the case that register tone languages lack contour
languages of Nepal. Tamang is characterized by four lexical tones, which tones. In fact, such languages frequently have rules of tonal assimilation
are referred to as 1, 2, 3, and 4. Whether a word in Tamang consists of one (termed "spreading") by which rising and falling ton_es are derived (see 6.2.3.1).
two, or three syllables, it is assigned only one of four contrastive tones' Other register languages have contour tones wh1ch are the result of two
However, as seen from the following comparison of these tones on mono: morphemes coming together. In Hausa, for instance, we find a falling tone
and bisyllabic words, in one of the two future tenses:
MONOSYLLABIC BISYLLABIC
na: z6: 'I will come' mwa: zo: 'we will come'
tone 1 ['""] ['-.,.] ka: 'you (m.)'
tone 2 [-] kyB.: 'you (f.)' kwa: 'you (pl.)'
['-]
tone 3 [-] [..r-] ya: 'he'
ta: 'she' swa: 'they'
tone4 [...,J [_{']
there is not always a perfect one-to-one correspondence in pitch between If we compare the form of these subject pronouns in the past tense,
a given tone on a monosyllabic vs. a bisyllabic word. Thus, although tone 4
is realized on a monosyllable as a L tone (which falls in utterance final na:z6: 'I came' mun z6: 'we came'
position), on two syllables it is realized as a L followed by a falling tone from ka: 'you (m.)'
kin 'you (f.)' kun 'you (pl.)'
H to M, that is, L-HMP In addition, Mazaudon (66, 82-84) points out that ya: 'he'
associated with these different word-tones are different states of the glottis ta: 'she' sun 'they'
(glottalization, breathiness, etc.), as well as different degrees of duration. She
considers-and convincingly argues against-various alternatives to recog- we see that it is possible to recognize a future marker fa/ with L tone, which
nizing the word as the unit of tonal representation. Thus, it is not possible combines with the underlying form of the subject pronouns, for example,
to assign an individual tone to each syllable, to recognize a two-way tonal fki+a/ 'you (f.) will,' fmu+a/ 'we will.' Thus, these falling tones are better
contrast with a movable accent, or to assign tone only to the first syllable of analyzed as a H followed by a L which come together across a morpheme
each word (with a phonological rule or rules spreading each tone over a word). boundary.
Each of these alternative approaches would fail in one way or another to While most African tone languages with contour tones show evidence that
account for the tonal properties of Tamang. a R should be analyzed as a LH sequence on one syllable, and that a F should
6.2.2.2.2 Contour Tones vs. Sequences of Level Tones In drawing be analyzed as a Hl. sequence, there are a few exceptions. Thus, in the dialect
the difference between the two types of tone languages, Pike (1948) states: of Kru reported on by Elimelech (1973), there is a H, a L, and a F which can
Contour systems differ from register systems in a number of points: (1) The be analyzed as :HL However, in addition, there is a rising tone which begins
basic tonemic unit is gliding instead of level. (2) The unitary contour glides at the level of a H and rises to a "super-high" level, and there is a falling tone
cannot be interrupted by morpheme boundaries as can the nonphonemic com- which begins at this "super-high" level and falls to L. While these two tones
could conceivably be analyzed as HS and SL (where S = super-high tone),
12
In this section T-T (where T = tone) stands for two tones on two separate syllables,
this S does not exist as an independent tone. As seen in Pike's third point,
while f;T 2 stands for two tones on the same syllable.
218 Suprasegmental Phonology 6.2
this provides some evidence that we are dealing with a true contour
rather than with a "compound" tone consisting of two level tones on
syllable.
Arguments to the effect that all phonetic contour tones should be u.u,:uv'""
as underlying sequences of level tones have recently been presented in the
literature (Woo, 1969). Leben (1973a:l23-125) presents evidence which
suggests not only that Thai has segmental tone, but also that its contoui: ..
tones must be analyzed as sequences of level tones, that is, LH and HL
than R and F. Since it is languages such as Chinese which Pike refers to
contour tone languages, let us take a closer look at the contours which
found in Chinese dialects. The four tones of Mandarin are specified pwv ........~.
ally as follows (Chao, 1965: 33):
tone 1 tl 55
tone 2 [ .... ] 35
tone 3 [v] 214
tone 4 [ l J 51
That is, tone 1 is realized on a high level pitch (5) , tone 2 rises from a M
level (3) to H (5), tone 3 falls slightly (from a 2 level to a 1 level) and then
rises almost to a H (4 level), and tone 4 falls from H (5) to L (!). While Pike
and most other tonologists have the intuition that the tones of Chinese are
best seen as single contour units rather than sequences of levels, Chao's
notation offers an important insight into the workings of tonal assimilations
in Chinese.
In Mandarin there is a tonal assimilation which takes place as follows
(Cheng, 1973:44): "In fast conversational speech, a second tone becomes
first when preceded by first or second tone and followed by any tone other
than the neutral tone." In terms of Chao's notation, this rule can be formal-
ized as follows:
35 -+ 55 f { ;~} _ T (where T = any tone except neutral)
That is, a 35 tone goes up to 55 (level H tone) when preceded by a tone which
ends in H (5) tone. A second rule of assimilation is now presented, from
Cantonese, which can be formalized as follows: 13
53 ->55/_ gn
In Cantonese a HM (53) falling tone becomes a H (55) tone when followed by
a tone which begins with a H (5) tone. If we were to state these two rules by
13
Both of these rules are discussed by Mohr (1973).
6.2 Suprasegmental Phonology 219
of indivisible contours, the following two rules would result for
~andarin and Cantonese, respectively:
R~ H/ {~}- T
F ~ H/- {~}
As written, these rules suffer from several shortcomings. First, while a R
becoming a H after a H may be viewed as assimilatory, no explanation is
given as to why R becomes H after H rather than before H. Similarly, no
explanation is given as to why F becomes H before H rather than after H.
Second, no explanation is given of why R should become H after R, or why
f should become H before F. Finally, using units such as Rand F fails to
reveal that exactly the same assimilatory process is responsible for both rules.
That is, if we were to write these rules using Chao's number notations, in
both Mandarin and Cantonese we would find that a 3 level rises to a 5 level
whenever it is wedged between two 5 levels, that is,
535 .... 555
In this formulation all of the above shortcomings are avoided. 14 Of course,
while this line of argument supports the division of contour tones into
sequences of phonetic pitch levels, it still may be the case (subject to verifi-
cation) that contour tones could represent an indivisible unit on a more
abstract level.
6.2.2.2.3 Distinctive Features of Tone The first attempt in the
literature to provide distinctive features of tone is Wang (1967). Closely tied
to the question of the kind of features necessary to capture tonal contrasts in
languages are those seen in 6.2.2.2.1 and 6.2.2.2.2, that is, should such
features be on segments or on suprasegments and should there be contour
tones. In addition, before providing features of any kind, one must establish
the realm of possibilities for tonal contrasts. Thus, a limit must be established
on the number of possible contrasting tone levels in any given language;
similarly, limits must be placed on the number of contrasting contour tones
(for example, how many rising tones can a language have?) as well as the
number of ups and downs possible on any given tone unit (whether the
segment or the syllable).
In general, the features proposed to capture contrasts betwee~ different
levels of tone mirror those that have been used to capture different vowel
heights. Thus, a contrast between H and L in a two-tone language is captured
14 The one remark that must be made is that in Cantonese, while S3#5 becomes 555,
5#35 does not become *5#55. Thus the exact position of the internal boundary is impor-
tant in stating the assimilation.
220 Suprasegmental Phonology 6.2
by calling the first [+High] and the second [ -High]Y Given a .......5 ...c:t~e<
with a three-way tonal contrast between H, M, and L, it is possible to use .
the features High and Low (mirroring the features High and Low used for
vowel height; see 2.4.4.2.1), or the features High and Mid. The latter features
are proposed by Wang (1967:97), the former by Sampson (1969:62-63):
WANG SAMPSON
H M L H M L
High + High + illustrates the use of these features:
Mid + Low +
In Sampson's feature notations, M is designated as sharing one property
with both H and L, since both H and M are [-Low] and both M and L are
[-High]. In Wang's system, M is classed with L, since MandL share a
[-High] specification, while Hand M share no feature in common. Should
a three-tone language reveal a functional similarity between H and M,
rather than between M and L, it would of course be theoretically possible
to specify M as [+High, +Mid], since it would still be distinct from H.
Languages have been reported with four underlying level tones, as well as
five in the questionable case of Trique (Longacre, 1952). For a language
with the four tones H, M, 'M, L (where 'M indicates a lowered-mid tone),
Wang's features High and Mid can be redistributed as follows:
H M 'M L 6.2.2.3 Natural Tone Rules Two recent studies (Hyman, 1973b;
High + +
Mid + +
Alternatively, another feature, which Wang calls Central, can be introduced,
which could also be used in the event that a clear case is made for five under-
lying tone levels in any language. In any event, other features such as Hight>
High 2 , Extreme, Raised, Lowered have been proposed, as well as features
intended to capture the relationship between tones and certain consonant
types (see 6.2.2.5). Note, finally, that Maddieson (1972:960) argues from
numerous African languages that different tone features should be used
depending on the phonological nature of otherwise identical phonetic con-
trasts. Thus, extending the notion of markedness discussed in 5.1.2ff to tone,
he argues that the same contrast between H and L may be analyzed as
[+Raised] vs. [-Raised] in one language, but as [-Lowered] vs.
[+Lowered] in another, depending on whether H or L is the marked tone
in the language.
Finally, Wang (1967) also gives the contour features Rising, Falling, and
Convex. A rising tone will of course be [+Rising], while a falling tone will
15
In the literature tone features are sometimes written in capitals (e.g. [+HIGH]) to
distinguish them from vowel height features such as [+high]. In this chapter they are
written with an initial capital.
6.2 Suprasegmental Phonology 211
be [+Falling]. A tone which, like Mandarin tone 3, first falls and then rises
is specified [+Rising, +Falling], while a tone which first rises and then falls
is specified [+Convex]. Superimposed on these contour features are the
features, High, Central, and Mid. Thus, [+High, +Rising] designates a
high rising tone (for example, the 35 second tone in Mandarin), while
[+High, +Falling) designates a high falling tone (for example, the 53 tone
of Cantonese). The following formalization of the two tone rules discussed
from Mandarin and Cantonese in the last section, taken from Mohr (1973),
+H~~h ]
[ +Rismg -+ [-Rising] I (+HFigh. ] _[-Neutral]
- a11mg
[ +Falling
+Hig~ ] -+[-Falling]/- (+H!~h
- R1smg
]
While these rules work, they are as unrevealing as the rules written with H,
R, and F in the previous section. Thus it should be clear that if tone features
are to reveal generalizations not captured by listing tones as units, much
work will have to be done in this area. In particular, it will have to be shown
that contour features are absolutely necessary and that level tone contrasts
should be captured by binary features. At present, the evidence for both is
inconclusive.
Hyman and Schuh, 1974) have provided typologies of natural tone rules and
have proposed various universals concerning the nature of these tonal
processes. A distinction is drawn between natural tone rules which have a
phonetic motivation and natural tone rules which have a grammatical basis.
These will be referred to, respectively, as phonetic and morphophonemic tone
rules.
6.2.2.3.1 Phonetic Tone Rules The two kinds of phonetic tone rules
which will be considered here are assimilation and simplification.
6.2.2.3.1.1 ASSIMILATION Like rules involving segments, a tonal
assimilation can be either anticipatory or perseverative. In addition, tonal
assimilations group themselves according to whether the assimilation is
vertical or horizontal. In a vertical assimilation, tones are raised or lowered
in the environment of a higher or lower tone. In an anticipatory vertical
assimilation, a tone is typically raised before a higher tone. Thus, Mbui has
a rule by which L is raised to M before H:
L-. Mf_H
As a result, underlying /niblli:/ 'breast' is realized as [nibw:]. In a perse-
verative vertical assimilation, a tone is typically lowered after a lower tone.
Thus, by the following rule,
H-+M/L-
111 Suprasegmental Phonology 6.2 6.2 Suprasegmental Phonology 113

~ H can be lowered to a M after a L. While this lowering process may some-. ltP' it is seen that a L can spread over an entire subsequent H-tone syllable
times involve a horizontal assimilation termed spreading (see below) the onlY when this syllable is in turn followed by another H, as seen in the follow-
~ollowing Gwari examples show that after L, H becomes M and M bec~mes ing Kikuyu derivation (Pratt, 1972:335):
M:
fgbr/ + /irs/ -+ [gorirs] 'bought' (immed. past)
/gyiwye da/ ~ [gyiwye da] possessor of money'
/jaaki da/ ~ [jaa 'ki da] 'possessor of donkey' Similarly, in b, a H can spread over an entire subsequent L-tone syllable
only when this syllable is in turn followed by another L. As shown by Hyman
Howeve:, altho.ugh the Mbui and Gwari examp!es show vertical raising '{l973b:l57-159) and Hyman and Schuh (1974:98), complete horizontal
and lowenng, as m a L-H sequence, vertical assimilations generally do not assimilation normally involves a telescoping (see 5.2.6.1) of two separate
occur when a preceding tone is higher than a following tone, as in a H-L processes, spreading (as seen in the partial horizontal assimilations above)
sequence. This fact is represented in the following table:
and absorption, as seen in the following rules:
Vertical Assimilation
L'HH~LH
NATURAL UNNATURAL
HLL ~ HL
L-H ~ M-H H-L-+ H-M
L-H ~ L-M H-L-+ M-L Absorption takes place when a contour tone is followed by a tone which is
identical to the end point of the contour. Thus, a LH rise becomes L before a
Finally, a third possibility for vertical assimilation in a L-H sequence is that
the L may rise as high as the His lowered. In this case we obtain a M-M H tone, and a HL fall becomes H before a L tone. The two steps involved
sequence (Meeussen, 1970). in complete horizontal assimilation are therefore represented as follows:
Horizontal assimilations result from a nonsynchrony between the tones
LHH~LLHH-+LLH
and. the segm~n~s \syllables) over which they have domain. In a partial
honzontal assimilatiOn, a contour tone, either rising or falling, results, as HLL~HHLL~HHL
seen in the following typical rules: First a contour is created by spreading, and then this contour is simplified by
LH-+ LLH absorption.
HL-+HHL While we have seen that vertical assimilations can be either anticipatory
or perseverative, horizontal assimilations are nearly always perseverative.
A L-H sequence may become a L-LH and a H-L sequence may become a This fact is represented in the following table:
H-HL, where LH represents a rising tone from L to H, and HL a falling tone
Horizontal Assimilation
from H to L. Examples from Gwari illustrating these two assimilations are
seen below: NATURAL UNNATURAL

/okpti/ ~ [okpa] 'length' L-H ~ L-LH L-H-+ LH-H

/sUkNu/ ~ [s6kU] 'bone'
In th~se examples, the tone of the first syllable spreads into the second syllable,
creatmg a contour tone. This spreading process is not complete, since a
trace of the second H-tone syllable remains in the LH rise, and a trace of
the second L-tone syllable remains in the HL fall.
Complete horizontal assimilation occurs when there is no remaining
phonetic trace of the underlying tone of the syllable onto which spreading
has occurred. Complete L-and H-spreading are seen in the following rules:
a LHH~LLH
b HLL~HHL
H-L ~ H-HL H-L-+ HL-L
Thus, we do not expect a L-H sequence to become a LH rise followed by a H,
nor do we expect a H-Lsequence to become a H~L fall followed by a L.
Spreading normally takes place in a perseverative fashion.
6.2.2.3.1.2 SIMPLIFICATION Simplification is the term used to refer
to rules by which contour tones are simplified to level tones. The Mandarin
rule by which the 35 rising tone becomes a H, and the Cantonese rule by
which the 53 falling tone becomes a H are examples of simplification. The
process of absorption cited in the preceding section also can be viewed as
simplification. In general, tone languages tend to level out contours, though
224 Suprasegmental Phonology 6.2
we have seen counteracting assimilations by which new contours are
duced. In the following derivation it is seen that horizontal assimilation
also apply to contours:
LHL-+L:HL-+LH
A LH rise followed by a L becomes, by spreading, a L followed by a RL
By a second horizontal movement, the L of the final lit fall is dropped at
end of the word. While languages have simplification processes operating
both rising and falling tones, rising tones seem to be less tolerated. Thus
context-free conversion of all LH rising tones to R is attested in certain
languages, for example, Hausa (Leben, 1971a:203).
6.2.2.3.2 Morphophonemic Tone Rules In addition to phonetic
rules of tonal assimilation and simplification, tone languages are charac-
terized by numerous grammaticalized rules. These all have in common that
they refer to specific morphemes or constructions.
6.2.2.3.2.1 DISSIMILATION Just as dissimilation most frequently is
bound to certain morphemes or constructions (see Johnson, 1973), the same
is true of tonal dissimilations. The only completely general tonal dissimilation
which comes to mind is reported by Leben (1971a:202). As seen in Leben's
formalization, in Rausa
L L ## -+ L H ##
[+long] [+long]
a L-L sequence dissimilates to become L-H when the vowel of the
L-tone syllable is long, and when this syllable is in word-final position.
Thus, underlying /karlmta:/ 'to read' is pronounced [karanta:]. With this
rule of L-tone dissimilation, Leben is able to explain a number of apparent
anomalies in the tonal structure of Rausa (see 6.2.2.3.2.3).
6.2.2.3.2.2 COPYING Copying refers to the process by which a
syllable (most frequently a grammatical morpheme such as a pronoun) is
considered to have no underlying tone of its own, but rather receives its tone
from a neighboring syllable. In Kru, the relative clause marker fa/ takes H
tone after a H-tone verb, and L tone after a L-tone verb. Since its tone is
always identical to that of the verb stem which immediately precedes it, this
morpheme is represented with no underlying tone. A rule of tone copying
will assign it the correct phonetic tone.
6.2.2.3.2.3 POLARIZATION As in the case of copying, rules of tone
polarization assume a morpheme with no underlying tone. In this case,
however, the morpheme is assigned a tone which is opposite to that of the
neighboring syllable from which it gets its tone. In Rausa, direct-object
pronouns are polarized with respect to the tone of the preceding verb, for
example, [mun ka:ma si:] 'we seized it' vs. [mun saye: si:] 'we bought it.'
6.2 Suprasegmental Phonology 225
In the first phrase /'Si:/ takes H tone, since the verb 'to seize' ends in L t~ne;
in the second phrase, fsi :/ takes L tone, since the verb 'to buy' e~ds m H
tone. The rule of dissimilation mentioned in 6.2.2.3.2.1 follows this rule of
polarization, as seen below:
fmUn. karanta: si/-+ mUn. karanta: si: -+ [mun karanta: si:] 'we read it'
The L-tone dissimilation raises the last syllable of /karanta:/ to H only after
. the pronoun I si :/ is polarized to this L' becoming a H tone. If the t~o rules
were to work in the opposite order, the wrong result would be obtamed:
fmun karanta: si:/-+ mun karanta: si:-+ *[mun karanta: sl:]
... Thus the only exception to the polarization of direct-object pronouns is
explained by Leben's rule of L-tone dissimilation.
6.2.2.3.2.4 REPLACEMENT By replacement is meant the process by
which the inherent tone of a morpheme is replaced by a grammatical tone.
Thus, in Igbo, the imperative is usually constructed by replacing the first
syllable with L tone and adding a suffix:
/ri/ 'eat' -+ [ri-e] 'eat!'
Tone replacement frequently takes place in the verb paradigm and in noun-
noun compounding. For example, in Mandarin, all four tones are replaced by
the so-called "neutral" tone in noun compounding (Cheng, 1973: 54ft').
6.2.2.3.2.5 FLOATING TONES In many cases where one might be
tempted to write a morphologized rule of tone alternation, an underlying
tone can be posited which has no underlying segments. Compare, for example,
the following realizations of the phrase 'jaw of monkey' in two Igbo dialects:
Central Igbo : [agba] + [eiJwe] -+ [agba eiJwe]
Aboh Igbo : [~gba] + [eiJwe] -+ [~gba eiJwe]
In both cases there appears to beaR-tone influence between the two nouns.
Instead of writing a rule by which L becomes R in possessive constructions,
an underlying R tone marker 'of' is recognized, as in the following under-
lying forms (see Voorhoeve, Meeussen and de Blois, 1969; Welmers, 1970):
Central Igbo : fagba ' eiJwe/
Aboh Igbo : /~gba ' eiJwe/
In Central Igbo this "floating" H tone is assigned to the left, while in Aboh
Igbo it is assigned to the right. Such :floating tones often explain otherwise
baffling tonal modifications which occur when words and morphemes are
strung together.
6.2.2.4 Terraced-Level Languages A number of African languages
exhibit tonal properties which prompted Welmers (1959: 3) to distinguish
between discrete-level and terraced-level tone systems. In the former, each
226 Suprasegmental Phonology 6.2
..toneme" is restricted to a relatively narrow pitch range and there is
no "phonemic overlapping" (see 3.2.1). That is, given a three-tone ""'-'&U<lllfloo
with H, M, and L, H will be higher than M and M higher than L anywhere
in the sentence. To illustrate this, Welmers presents the following Jukun
sentence meaning 'who brought these yams?":
/ani ze SUra a syi nf bif -+ r- _- __ ---]
In numbers we could represent this as 3-2-1-3-1-1-2-3-2.
6.2.2.4.1 Downdrift In many African languages, on the other hand,
a sequence H-L-H is not realized as [- _ -], but rather as [- _ -].
That is, the two H tones are not pronounced on the same pitch level (though
they are phonologically identical), but rather the second H is lower in pitch
than the first. Stated differently, the interval from H to Lis greater than from
L to H. This phenomenon, known as downdrift, applies progressively to each
H preceded by a L, as seen in the following Igbo sentence:
? na auwa Jnya fgwe ' he is trying to ride a bicycle'
HLHLHLHL
- -_]
In the above example, the downdrifting effect extends over several H-L-H
sequences. As shown in phonetic brackets, the L tones which intervene
between the H tones are also subject to downdrift, though the degree of
lowering varies from language to language. In Hausa, for instance, a H late
in a sentence can downdrift to a pitch level which is phonetically lower than
a L which appears early in the sentence, as seen below:
Ba Ia: da She: hU za: su z6: 'Bala and Shehu will come'
LHLHLHLH
[- _-]
If we were to assign pitch integers to the different tones, we would have a
sequence 4-6-3-5-2-4-1-3. Thus, the 4 of the initial L is higher than the
3 of the final H. Numerous formalizations of downdrift have been devised
to assign such integers (Schachter and Fromkin, 1968: 108; Voorhoeve,
Meeussen and de Blois, 1969:82; Carrell, 1970:98; Williamson, 1970;
Fromkin, 1972:56--57; Schadeberg, 1972; Peters, 1973; for theoretical
discussion, see Stewart, 1971). What is consistent in the above integers is
that Lis always two steps below the last H. Also, H 2 in a HcL-H 2 sequence
will be realized one step below Ht> and L 2 in a LcH-L 2 sequence will be
realized one step below L 1 The assigning of an underlying tone will therefore
not be done on the basis of absolute pitch; rather, it will be done on the basis
of the relationship of a given phonetic pitch to surroU'lding pitches.
6.2 Suprasegmental Phonology 21.1
6.2.2.4.2 Downstep While downdrift represents an automatic
lowering process (see, however, 6.2.2.4.3), a lowered H receives phonemic
status when a L which "conditions" downdrift is lost (either through
deletion or through assimilation). The standard example comes from Twi
(Fromkin, 1972:57):
/mi ~bu/ 'my stone'
pitch-assignment: 31 3
downdrift: 2
vowel deletion: 0
[mf 'bu]
3 2
First the integers 3 and 1 are assigned to H and L, respectively, in such
phrases (see Fromkin, 1972; Peters, 1973 for more detailed discussion).
By downdrift, the 3 of the second H is lowered to 2. At this point a rule of
vowel deletion deletes j:,j, and the result is a 3-2 sequence, that is, a H tone
followed by a downstepped 'H. Since on the surface we now have a phonetic
contrast between H-H, H-L, and H-'H, a new "toneme" has come into
existence.
Many of these downsteps can be predicted morphophonemically, as in the
above example. Others, however, cannot be, and must be treated as a third
tone. As pointed out by Welmers (1959: 3), it would be mistaken to call this
0
tone a M, since this tone does not contrast with H after L (see Tadadjeu,
1974 for a counterexample from Dschang-Bamileke). More important,
however, in languages with true M tones, a sequence H-M-H is realized
[ - - -], that is, 3-2-3, with the second H rising above the level of the
preceding M. In a language such as Twi, however, a H-'H-H sequence is
realized as [- - -], that is, 3-2-2, with a following H realized on the same
pitch level as the preceding 'H. In other words, a downstepped high tone
establishes a terrace just like a regular H tone, and no tone can go higher
than this ceiling. For this reason, Welmers refers to such languages as
terraced-level.
6.2.2.4.3 Intonation and Tone The relationship between downdrift,
representing an automatic lowering process, and downstep, representing a
nonautomatic phonemic tone, is now generally acknowledged (see Stewart,
1967, 1971). While recent studies such as Voorhoeve (1971) and Tadadjeu
(1974) have shown that downdrift is not a necessary prerequisite for down
step (compare Meeussen, 1970), most cases of the latter do in fact derive from
the former.
A relationship which is not as well understood is that between intonation
and tone. Schachter (1965) argues convincingly that downdrift is an into-
national property, since in languages such as Hausa it can be suspended for
purposes of emphasis or question. Virtually all tone languages exhibiting
228 Suprasegmental Phonology 6.2
automatic downdrift have only two tones, H and L. Most dialects
Yoruba, Nupe, Ewe, and Jukun, all ofwhich have H, M, and L, do not
downdrift. An explanation for this has recently been proposed by Homtertc
(1974). Hombert shows that if a three-tone language were to let the
of a H-L-H sequence undergo downdrift, it would be likely to be COIUW;t\lf
with an underlying M (compare LaVelle, 1974). Hombert further shows that
intonational lowering can, in other sequences, also be accounted for by
reference to the tonal contrasts of a language. In most African language$, .
for instance, a sequence H-H-H is realized as [ - - -] rather than as
[ - - -]. That is, sequences of H tones are realized on the same pitch level. . .
The reason is that in these languages H-H contrasts either with H-M or
H-'H, and perceptual confusion would result if H-H were to undergo
lowering. In languages such as Hausa and Shona, where there is no M and
where tone is less important for lexical contrasts, sequences of H tones
do in fact lower (see Meyers, 1974). Finally, sequences of L tones almost
always descend in pitch; only Dschang-Bamileke (Tadadjeu, 1974) has
a contrast between L-L and L-'L which could be potentially confused. 16
The conclusion is that it is to be expected that intonational lowering will
occur, except where there is a tonal contrast which would be obscured. For
further discussion, see Hombert (1974).
6.2.2.5 Consonant Types and Tone While the tone rules of the
preceding sections have been presented without reference to segmental
information, different consonant types frequently interact with natural tonal
assimilations. In Nupe, as seen in the following forms,
/paJ 'peel' [epa] 'is peeling'
/bel/ 'be sour' : [eba] 'is sour'
/waJ 'want' : [ewa] 'is wanting'
L-spreading takes place only when the intervening consonant is voiced. In
Ngizim, on the other hand, H-spreading takes place when the intervening
consonant is voiceless (for example, /p/), a sonorant (for example, fwf) or
an implosive (for example, fo/). In other words, certain consonant types
are more amenable to L-spreading or H-spreading operating through them.
As proposed in Hyman and Schuh (1974:108), a voiced obstruent can block
the spreading of a H tone through it, just as a voiceless obstruent can block
the spreading of a L tone through it. Sonorants are neutral with respect to
tone, since they allow both L and H to spread through them.
Such examples which are numerous in African tone languages (for example,
16
According to recent investigations by the author and Jean-Marie Hombert, a phonetic
fall is the pnmary perceptual cue for low tone. This fact is at least in part responsible for
the phenomenon of downglide (Stewart, 1971: 185) by which a Lin utterance-final position
is realized as a marked fall in many languages.
6.2 Suprasegmental Phonology 229
Ewe Xhosa, Shona) point to the generalization that consonant types affect
tone'but tone does not affect consonant type~._While a voiced o~stru~nt shows
an affinity for L tone, L tone does not votce consonants. This pomts to an
mportant difference between stress and tone, since we saw in 6.2.1.2.3 that
~tress has many effects on segments. While a number of explanations have
been proposed to capture the relationship between voicelessness and H tone
on the one hand and voiced obstruents and L tone on the other, none of
these has received universal acceptance by phoneticians. Halle and Stevens
(1971) and Halle (1972:181) propose to capture this relationship by means
of the features Stiff Vocal Cords vs. Slack Vocal Cords, as follows:
VOWELS OBSTRUENTS SONORANTS
vv v p b w
stiff - - + + -
slack + - - - +
H tone and voiceless obstruents share stiff vocal cords, while L tone and
voiced obstruents share slack vocal cords. Both M tone and sonorants
represent the neutral state of the vocal cords. Another set of distincti:e
features based on larynx height is proposed by Maran (1971 :14), whlle
others emphasize the rate of air flow through the glottis as the primary
factor responsible for this interaction. (For a collection of papers dealing
specifically with the topic of consonant types a~d tone, ~ee Hyman, 1~7~c.)
Since voiceless and voiced obstruents have dtfferent pttch charactensttcs,
a tonal contrast can reconstruct as an earlier voice contrast. Thus, Mandarin
35 reconstructs as an earlier H (55) tone with an initial voiced obstruent.
This voiced obstruent lowers 55 to 35 and then devoices. If the first tone
contrast in a language can be traced back to a voicing contrast, one speaks
of tonogenesis (Matisoff, 1973 :73).
6.2.3 Typologies of Prominence
In the preceding sections, stress and tone have been treated as two
diametrically opposed types of prominence. Some of the differences between
stress and tone systems have already been alluded to. These differences are
summarized as follows:
1. In a stress language prominence is culminative; in a tone language
prominence is nonculminative. Thus, while only one syllable per word can
have primary stress, any number of syllables in a word can have H tone,
subject to the sequential constraints of the language.
2. In a stress language prominence is syntagmatic; in a tone language
prominence is paradigmatic. Thus, while in a stress language one syllable in
a word is singled out for stress, each syllable of a tone language receives
230 Suprasegmental Phonology 6.2 6.2 SuprasegmentaJ Phonology 231

tone, often choosing from a number of contrasting values (for example, basis ofthis misunderstanding. The term musical, on the other hand, indicates
H, M, and L 1 7) or kinds of prominence. that it is a tone which is assigned culminatively to a given syllablewithin a
3. In a stress language we find rules of stress reduction; in a tone language word. Languages such as Swedish and Serbo-Croatian, for instance, have
we find rules assimilating and dissimilating tones (see McCawley, 1964, contour tones assigned to words. In Serbo-Croatian there is both a rising and
1970). a falling tone, and either of these can be long or short. This means that in
4. In a stress language, presence vs. absence of stress can condition seg- addition to the placement of this "musical accent," speakers must pay
mental changes (for example, diphthongization under stress, vowel reduction attention to the direction of the pitch change, since rising and falling tones
under stresslessness); in a tone language, consonants typically affect tone, contrast on potentially the same syllable. Prominence is still culminative,
rather than the reverse. since only one such tone can be assigned per word. In more recent terminology,
There are other criteria which are also sometimes used to type different these languages would be called pitch-accent, as opposed to stress-accent.
systems of prominence. Voorhoeve (1973), for instance, focuses on the In a stress-accent language, a single culminative mark of prominence is
difference between lexical and rule-governed prominence. In 6.2.1.1 a possible on a given syllable of a word. The perceptual cues of this stress can
distinction was made between free and fixed stress. If stress is free, that is, be changing pitch, vowel duration, or greater intensity, all contributing to
unpredictable, falling on the first syllable in some words and on the second the highlighting of the stressed syllable. In a pitch-accent language, prom-
in others, then its exact position must be part of the lexical entry for each inence is assigned to a given syllable of a word, but there can be two or more
word. If stress always falls on the same syllable (for example, initial or kinds of prominence (for example, a rising vs. a falling contour). Pitch-accent
penultimate), stress need not be a part of the lexical makeup of underlying languages are thus tonal to the extent that the feature which is assigned is
forms. Stress languages can be of either type, or even intermediate, with tone (and that this tone can contrast with another tone in the same position).
stress being partially free, partially fixed. Tone languages, on the other hand, Pitch-accent languages are like stress-accent languages, however, in that
are normally assumed to have tone indicated as part of the lexical item. In there cannot be more than one syllable per word which receites the tonal
this typology a language such as Russian, which has unpredictable stress, accent; that is, prominence in pitch-accent languages is culminative.
would be grouped with tone languages such as Thai or Y oruba, since each of For a language to be called pitch-accent, it is, however, not necessary for
these languages would require some indication of prominence in the lexicon. there to be a tonal contrast. Thus, Voorhoeve (1973) for Safwa and Schade-
The question of determining an adequate typology of prominence has berg (1973) for Kinga show that in these languages there is normally only
received considerable attention from a number of linguists (Pike, 1948; one H tone per word. In Japanese, as treated in great detail by McCawley
Welmers, 1959; Martinet, 1960; McCawley, 1964, 1968, 1970; Woo, 1969; (1968), each word can be treated for prominence by indicating the placement
Voorhoeve, 1973). While stress and tone represent the logical dichotomy of a "pitch fall." The following accentual possibilities for words with one,
within such typologies, it is quite clear that many languages fall in one respect two, and three syllables are represented in Table 6.1 (McCawley, 1968:132).
or another midway between stress and tone. First, it is quite clear that stress
exists in at least some tone languages. We have already referred to Mandarin
Table 6.1 Accentual Patterns on Japanese Words of 1, 2, and 3 Syllables
Chinese (see 6.2.1.2.3), where the neutral tone results from the lack of stress.
In many Bantu languages which are tonal (for example, Shona), there is in Underlying Accent Pitch Following Pitch
addition a superimposed penultimate stress which lengthens the vowel of
/hi/ 'fire' [-] [-]
this syllable. Stress and tone are therefore not mutually exclusive (see Woo,
/hi/ 'day' [-] [-]
1969; McCawley, 1970).
6.2.3.1 (Dynamic) Stress vs. Pitch-Accent (Musical Stress) One of /s6ra/ 'sky' [ - _] [-]
the dichotomies drawn in Prague studies of prominence (for example, /kawaj 'river' [--] [-]
Trubetzkoy, 1939; Jakobson, 193la) is that between dynamic and musical [-]
[-- 1
stress. Dynamic stress is what we referred to as stress in 6.2.1. While force of
articulation and the resulting intensity of the speech signal are not necessarily
/take/ 'bamboo'
/kabuto/ 'helmet' [- __ ] [-]
the major perceptual cues of stress, the term dynamic was chosen partly on the /kok6ro/ 'heart' [_-- 1 [-]
f?otok6/ 'man' (_--] [- 1
17 The existence of such word-tone languages as Tamang (see 6.2.2.2.1) should, however, /katati/ 'form' [_ --1 [-]
be borne in mind.
232 Suprasegmental Phonology 6.2
In the forms in the table, it is observed that there is always one and only one
pitch fall, which can be realized within the word or on the following syllable
of the next word (or suffix). Since there is a rule by which a H-H pitch
sequence at the beginning of a word in Tokyo Japanese is converted to
L-H, we can recognize the following intermediate possibilities:
monosyllabic words: H + (L)
H +(H)
bisyllabic words: H-L + (L)
H-H + (L)
H-H +(H)
trisyllabic words: H-L-L + (L)
H-H-L + (L)
H-H-H + (L)
H-H-H +(H)
As seen in the underlying forms o~ !he-table, the pitch contours of Japanese
words can be predicted by placing an accent /'I on the vowel which imme-
diately precedes the pitch fall. If there is no pitch fall within a word, either
the last syllable is accented, in which case the fall will be realized on the
suffix syllable, or there is no accented syllable, in which case a suffix syllable
will be realized without an accentual fall.
It should be quite clear that although we are talking about pitch and
pitch falls, Japanese is not a tone language. In fact, it differs from stress
languages only in that the accentual pattern is spread throughout the whole
word, rather than being realized phonetically on one syllable. There are, it
should be noted, stress languages which, like Japanese, have words without
any stress (see footnote 8). While it would be distorting the nature of Japanese
to speak of a H and a L on each syllable, Japanese should be compared with
such word-tone languages as Tamang (Mazaudon, 1973), which was discussed
in 6.2.2.2.1.
6.2.3.2 Monotonic vs. Polytonic Accent The term accent has
been used to refer to systems of prominence where the assigned feature is
culminative (either stress or tone). A second dichotomy made by JakQbson
(1931a) and Trubetzkoy (1939) is based on the number of contrasting cul-
minative tones found in a language. Stress languages such as English,
Russian, Japanese, and Finnish are monotonic since they assign only one
kind of culminative accent. Safwa and Kinga, which assign one H tone per
word, are also monotonic. On the other hand, languages such as Swedish,
Serbo-Croatian, and all tone languages are polytonic, in that a contrast
between at least two different kinds of prominence is possible in the same
position in a word.
While this dichotomy seems straighforward, it sometimes runs into
difficulties. In a language such as Greek, for instance, where there is a con-
6.3 Suprasegmental Phonology 233
trast between rising and falling accents only on a syllable with a long
vowel or diphthong, it is possible to decompose these contours and assign
stress to morae, as follows:
rising tone: CvV
falling tone: CVV
In this analysis, Ancient Greek is judged to be monotonic, and as a result
can be viewed as a stress language which assigns prominence to morae.
A mora can be defined basically as a light (that is, CV) syllable, or as each
of the two parts of a heavy (that is, CV-C or CV-V) syllable. It is quite clear
that if Ancient Greek is treated as having syllable prominence, it is then
polytonic (with rising and falling tones); if it is treated as having mora
prominence, it is monotonic. In the second case, the same division of syllables
into morae that was seen in connection with stress placement in 6.2.1.2.2 is
observed.
6.3 Other Suprasegmentals
While most studies of suprasegmental features center around the
various kind of prominence systems (stress, tone, pitch accent), some linguists
have attempted to view other phonological features as suprasegmental-at
least in some languages. The two features which will be briefly treated in this
section are vowel harmony and nasalization.
6.3.1 Vowel Harmony
By vowel harmony is meant that all vowels within a specified
(suprasegmental) unit agree in some phonetic feature. The question is whether
this feature should be interpreted as a property of segments or of grammatical
units larger than the segment (for example, stems, words).
6.3.1.1 Types of Vowel Harmony An attempt to provide a frame-
work for typologizing vowel harmony systems is provided by Aoki (1968).
Aoki first distinguishes between partial and complete vowel harmony. In
complete vowel harmony, which can also be seen as a kind of reduplication,
the vowel of a morpheme completely assimilates to another vowel. An
example can be found in certain central dialects of Igbo, where a verb such
as /me/ 'make, do' takes in the past tense the consonant /r/ followed by a copy
of the vowel of the verb stem, that is, mere 'made, did'; compare mara 'knew'
from fmaf 'know' plus /r/ plus a copy of the stem vowel fa/. A process of
vowel harmony occurred historically, since dialects in the Onitsha area
pronounce 'made' melu, revealing that the past tense suffix reconstructs
as *lu.
234 Suprasegmental Phonology 6.3
While complete vowel harmony is often referred to as vowel copying or
vowel reduplication, most cases referred to as vowel harmony are of the
partial variety. In this case a vowel assimilates in certain features to another
vowel. The most common features assimilated are front-backness, tense-
laxness and labiality. An example of front-backness harmony is found in
Hungarian (Vago, 1973: 581 ). The first person plural suffix 'we' is realized
as unk after back vowels and iink after front vowels, as seen in the following
forms:
hoz-unk 'we bring' iii-link 'we sit'
varr-unk 'we sew' ver-iink 'we beat'
However, as pointed out by Vago, there are certain verb stems with /i/ and
fi:f which exceptionally take back vowels in their suffixes, for example,
szid-unk 'we curse,' not *szid-iink. In order to predict the back vowel found
after such stems, we are faced with either recognizing these stems as excep-
tional (specifically by marking these forms with a rule exception feature
[-vowel harmony] which would prevent funk/ from becoming iink), or
with positing an abstract underlying high central unrounded vowel fi1 in the
stem szid. By a low-level phonetic rule all instances of ji/ would be converted
to [i], but only after vowel harmony had had a chance to apply. This second
solution brings us into the abstractness controversy (see 3.3.5), which is the
concern of Vago's study of vowel-harmony systems.
The second feature which is frequently found to be assimilated in vowel
harmony is tense-laxness. This feature has been treated as an opposition
between tense and lax, covered and noncovered (Chomsky and Halle,
1968: 314-315), and advanced vs. retracted tongue root (Stewart, 1967).
Thus, in Central lgbo two sets of four vowels are found (see 2.4.2.3):
ADVANCED TONGUE ROOT :RETRACTED TONGUE ROOT
u i 11
e 0 a Q
The vowels /i/ and N/ give the impression of a very tense closed [ eJ and [ o],
respectively, while /9/ resembles [:>].What is important is that in constructing
words in lgbo, all vowels found within# boundaries are chosen from one of
these sets. Thus there are words such as fe'go/ 'money' and fa'glf/ 'leopard,'
but no words such as *fe'g(lf and */a'go/. As a result, the verbal noun prefix
is pronounced [e] before the stem vowels /i, e, u, of and [a] before the stem
vowels /i, a, If, 9(, as seen in the following forms:
/sf/ 'wash' -+ [esi] 'washing'
jsj/ 'say' -+ (as!] 'saying'
The third feature found in vowel-harmony systems is roundness. In Turkish
(Zimmer, 1967, 1970) high vowels agree in both backness and roundness.
6.3 Suprasegmental Phonology 235
'fbUS, the vowel in the momentary suffix flyorf is pronounced [i] after front
unrounded vowels, [ii] after front rounded vowels, [i] after back unrounded
vowels, and [ u] after back rounded vowels, as seen in the following forms
(Zimmer, 1970:90):
(istiyor] 'he wants'
[soyliiyor] 'he is saying'
[anliyor] 'he understands'
[kutluyor] 'he is celebrating (some occasion)'
6.3.1.2 Approaches to Vowel Harmony From the above discussion
it is seen that vowel harmony applies to all vowels within a given domain
(normally between # boundaries). As with other phonological rules, vowel
harmony can be blocked by a strong grammatical boundary. Thus, when the
two Igbo verbs fgaf'go' and /fe/ 'cross' are compounded, the result is [gafe]
'go across' and not *[gaHt] or *[gefe]. It is assumed that the underlying
boundary in fga#fe/ blocks the application of vowel harmony. As this
boundary weakens to a +, vowel harmony may be able to penetrate it. Thus,
some speakers pronounce fbu#ta/ 'to carry (away)' as [Mta] (breaking
vowel harmony), while others pronounce it as [bute]. In the latter case,
#has weakened to + (see the discussion of boundaries in 6.1.2.2).
The question is whether vowel harmony is a suprasegmental or a segmental
property. When there is vowel harmony across a boundary, there is no need
to discuss the underlying representation of vowel harmony, since an affix
vowel can be seen to assimilate to the vowel in a neighboring syllable. When
the vowel harmony is within a morpheme it is not clear whether one should
speak of one vowel assimilating to the other or of a suprasegmental assign-
ment of the shared vowel feature. Thus, Finnish, which is characterized by
front-backness vowel harmony, has the two words [pofitii] 'table' and
[pouta] 'fine weather.' These words differ in that all of the vowels in 'table'
are [-back], while all of the vowels in 'fine weather' are [+back]. Within
the framework of generative phonology, there have been three approaches to
the underlying representation of vowel harmony (see Kiparsky, 1968a;
Vago, 1973). In the first, an underlying abstract feature such as [+Back]
and [-Back] is assigned to each morpheme (Lightner, 1965). In this case,
the two Finnish words would be represented, respectively, as fpoutafc-BackJ
and fpouta/[+BockJ In the second approach, one vowel (for example, the
first or last) is fully specified, while all other vowels in the same morpheme
are represented by means of archiphonemes, that is, partially specified
segments in the underlying form (see Bach, 1968; Carrell, 1970). In this case
the two Finnish words would be represented as fpoUtA/ and /poUtA/,
respectively. The archiphonemes /U/ and /A/, which are unspecified for
backness (see 3.2.2), are converted to [ii] and [a] after front vowels, [u]
and [aJ after back vowels. In the final approach, as argued by Kiparsky
236 Suprasegmental Phonology 6.3
(1968a), all vowels within morphemes are fully specified (see Stanley, 1967),
and the fact that all vowels agree in backness within a morpheme is captured
by means of a morpheme structure condition (see 4.2.1.2). In this last ap..
proach, the two Finnish words would be represented as fpouta/ and /pouta{.
Of the three approaches, only the first treats vowel harmony as a supra~
segmental property. In the second approach, vowel harmony is seen to be the
property of, in this case, the first vowel of each morpheme, while in the third,
it is seen to be a redundant property of morphemes. In all approaches, a
rule of vowel harmony assimilates vowels across morphological boundaries.
6.3.2 Nasalization
The case for analyzing vowel harmony as a suprasegmental property
has received less and less support; in contrast, recent arguments have been
put forth (Leben, 1973a,b) suggesting that nasalization may be considered
a suprasegmental feature in some languages. Both vowel harmony and
nasalization were seen to be "prosodic" in the British (or Firthian) school, as
evidenced, for example, by Carnochan's (1960) analysis oflgbo and Robins'
(1957a) analysis of Sundanese. In addition to Sundanese, nasalization takes
on a suprasegmental appearance in Terena (Bendor-Samuel, 1960), Desano
(Kaye, 1971), and Guarani (Lunt, 1973), although Langendoen (1968) pro-
poses a restatement of such phenomena without "prosodies." The three
languages receive close attention from Leben (l973a,b), who presents Terena
vowel nasalization as follows:
In forming the first person singular:
(a) Nasalize all vowels and semivowels in the word up to the first stop or
fricative
(b) Nasalize the first stop or fricative in the word as follows: mb replaces p,
nd replaces t, IJU replaces k, nz replaces both s and h, and ni replaces both s
and hy (1973a: 142-143).
Thus, the following oral-nasal opposition is found in comparing the third
person singular and first person singular forms:
emo?u 'his word' emo?il 'my word'
ayo 'his brother' ayo 'my brother'
owoku 'his house' owol)gu 'my house'
piho 'he went' mbiho 'I went'
ahya ?aso 'he desires' anza ?aso 'I desire'
It is quite clear from these forms that nasalization is the distinguishing
feature between third and first person singular, and that nasalization (or
orality) is realized potentially over several syllables. It is this latter feature
which suggests a suprasegmental analysis of nasalization in Terena.
In determining whether nasalization should be viewed as segmental or
suprasegmental for any given language, several factors must be considered.
6.3 Suprasegmental Phonology 237
First, can a directionality for nasal spreading be established? In the above
examples, nasalization clearly spreads from left to right. Hence it is possible
to recognize a nasal element 'first person singular' which is prefixed to
nouns and verbs, let us say /N/ (for example, fNemo?uf 'my word'), which
causes the perseverative spreading of nasalization. A later rule deletes /N/.
In such a fashion, one could avoid analyzing nasalization as an underlying
suprasegmental property. The counterargument to such an analysis for
Terena is that the exact phonological shape of this underlying nasal element
is indeterminate, since its sole specified feature is [+nasal]. The same
argument has been used against "floating tones" (see 6.2.2.3.2.5), which are
specified only for tonal features.
Since certain consonants (specifically nonlow obstruents) block the
spreading of nasalization, a directionality can be established. It is this
directionality which in turn provides the possibility of a segmental analysis
of Terena nasalization. In a language such as Desano (Kaye, 1971), on the
other hand, where nonlow obstruents also become nasalized (b becomes
m, d becomes n, etc.), a segmental analysis is much more difficult to maintain.
In Desano, morphemes are marked as a unit as either [+Nasal] or [-Nasal],
or are unspecified for nasality. Thus, the morphemes [ wai] 'name' and
[ wai] 'fish' differ in that the first is recognized as fwai/ with the feature
specification [+Nasal], while the second is recognized as jwai/ with the
feature specification [-Nasal]. Morphemes left unspecified for nasality are
typically those which become nasalized in the context of another morpheme
marked [+Nasal].
The motivation for recognizing a suprasegmental feature Nasal is seen
from the forms [Jlohso] 'kind of bird' and [yohso] 'kind of lizard.' Kaye
analyzes these as fyohsofc+Nasall and fyohsofc-Nasall respectively. We have
already said that nasalization is not blocked by nonlow obstruents. In
addition, unspecified morphemes become [+Nasal] both before and after
[+Nasal] morphemes, as seen in the following derivations (see Leben,
1973b:142):
/sedafr+NasatJ + /du/ -+ [senanil] 'pineapple'
/gof[-NasatJ + /du/-+ [goru] 'ball' 18
/bli/ + /dafr+NasatJ -+ [milna] 'old men'
/bli/ + /glih-Nasa!J -+ [bligli] 'old man'
Since an unspecified morpheme becomes nasalized on either side of a
[+Nasal] morpheme, it is not possible to mark nasalization on only the first
(or last) vowel of a morpheme and then copy nasalization throughout the
morpheme. The only way to avoid analyzing nasalization as an underlying
suprasegmental property is to follow Kiparsky's (1968a) suggestion for vowel
18 Underlying /d/ is converted to phonetic [r] in this position.
238 Suprasegmental Phonology 6.3

harmony and fully specify [+nasal] on each nasalized segment of the

called [+Nasal] morphemes, for example, /pohsof 'kind of bird.' While
such underlying forms will exhibit considerable redundancy, these redun~
dancies can be captured, as in the case of vowel harmony, by morpheme
structure conditions.
In summary, then, the issue of whether vowel harmony and nasalization
are suprasegmental in the same sense as stress and tone is as yet largely
unsettled. 19 ftPPE:NDIXE:S

List of Symbols
c -consonant v -vowel
ch
co
-aspirated v -nasalized
- unaspirated v: -long
cY - palatalized v -stressed (or high tone)
cw - labialized v -tense (or mid tone)
c: - long or geminate y - pharyngealized
<; - pharyngealized y -voiceless
N -nasal L -liquid
:t:T -voiceless
-syllabic
.
L
~
-voiceless
-syllabic
li
G -glide 0 - zero or null segment
G -voiceless
IAB/ - phonemic slashes {AB}- morphophonemic braces
[AB]- phonetic brackets ( + F] - distinctive feature brackets
*AB - unattested (either a historical reconstruction or a disallowed
u While duration (vowel and consonant length) is normally treated along with stress and sequence)
tone as a suprasegmental (see Lehiste, 1970), this topic will not receive specific attention
A --.. B f _ C - A 'becomes' by phonological rule B before C
here, since we have already had occasion to refer to length in conjunction with other issues
in phonology. A -. B j C _ - A 'becomes' by phonological rule B after C

239